Software-defined Defence:

Algorithms at War
Dr Simona R. Soare, Pavneet Singh and Meia Nouwens

February 2023

The International Institute

for Strategic Studies
Contents
Executive Summary  2

Introduction3

1. Conceptual Underpinnings of Software-defined Defence 5

1.1. The changing relationship between hardware and software in generating military advantage 5
1.2. Data-centric architectures and forces 6
1.3. Software as the core of modular weapon and network design 7
1.4. A human-defined and human-centric approach by design 8

2. A World of Innovation Microscalers 10

2.1. Late adapters to a fundamentally different dual-use technology landscape 10
2.2. Defence software and AI/ML investment 11
2.3. Not agile enough 16

3. Software-defined Defence and Sino-American Strategic Competition  19

3.1. US 19
3.2. China 23

4. European Approaches to Software-defined Defence 28

4.1. UK 29
4.2. France 33
4.3. Germany 37

5. Conclusion  40

Notes42

Cover
Cloud computing. (Just_Super via Getty Image)
Executive Summary

Software and artificial intelligence (AI) are critical ena- functionality of systems increasingly determines opera-
blers of modern military operations, lead the evolution tional advantage in information superiority. Secondly,
towards multi-domain operations, enhance interoper- software-defined defence requires a data-centric approach
ability among allied forces, and support the achieve- to developing new capabilities and systems-of-systems.
ment of information superiority and decision-advantage Thirdly, it takes a human-centric approach to design-
against adversaries. Much of the functionality and per- ing API-enabled end-to-end electronic workflows that
formance offered by military equipment, from the F-35 enhance human capacity and safety. Finally, software-
Lightning II fighter jet and the Patriot missile-defence defined defence regards advanced defence software and
system, to the M1 Abrams tanks and the French Griffon, AI/ML as a core weapon capability and therefore places
Jaguar and Serval armoured vehicles, is already software- emphasis on the software component in early system
defined. As software now drives most of many military design, as well as in subsequent upgrades.
platforms’ functionality, it is increasingly clear that it is The processes used in the development and deploy-
not merely layered on to military hardware. Software is ment of advanced defence software and AI/ML remain
part and parcel of a weapons system. embedded in decades-old, hardware-driven waterfall
This report investigates the growing role of defence capability-development models. Efforts to use agile, iter-
software and AI/ML (machine learning) in military power ative and DevSecOps frameworks are incipient across
now and in the medium term. It focuses on three goals: all five countries analysed. However, ongoing initiatives
 to define software-defined defence. The paper are slow and cumbersome, the causes of which are archi-
considers software-defined defence to be a fun- tectural, organisational and operational. Most advanced
damental architectural, organisational and opera- defence software is embedded in bespoke hardware,
tional principle of modern military operations. which requires modifications to add new software
Software-defined defence entails a new logic for functionality and improve performance, and defence
capability development which disaggregates sen- industries lack enabling digital infrastructure and suf-
sors from effectors, software from hardware, and ficiently skilled operators. Examples of core operating
data from specific applications, while connecting systems for capability families are slowly beginning to
them in data-centric, multi-modal, multi-domain, emerge in France, the UK and the US. But more work
adaptative battle networks; is needed to move towards a defence-as-a-platform and
 to assess ongoing practices and processes in the software-as-a-service-approach.
development of defence software and AI/ML, and As Sino-American strategic competition intensifies,
identify recurring challenges; with the integration of advanced technologies like AI/ML
 to explore and assess the ongoing efforts towards at its core, China’s investment in software-defined defence
software-defined defence in five country case will narrow the West’s military-power advantage. The US
studies – China, France, Germany, the United is racing to meet this threat and is consistently attempt-
Kingdom and the United States – and how Sino- ing to accelerate the safe and responsible integration of
American strategic competition is shaping them. defence software and AI/ML into its defence capabilities.
Similar efforts are only incipient and at smaller scale in
Software-defined defence is based on four foundations. France, Germany and the UK. A transatlantic software-
Firstly, a changing relationship between military soft- defined defence gap has already emerged – one which
ware and hardware, in which technological progress is could still be bridged if Europeans choose to embrace
faster in software than in hardware, and software-defined defence software and the digitalisation of defence.

Software-defined Defence: Algorithms at War 2

Introduction

Software and artificial-intelligence algorithms are force software and AI/ML adoption, and considers specific
multipliers, enhance operational art and troop morale, examples to demonstrate the bottlenecks and obstacles
and contribute to victory and defeat on the battlefield. to more effective adoption of modern defence-software
Software not only underpins modern military capabili- solutions. Wherever possible, the report uses examples
ties but also enables them to collect and analyse data of military capabilities shared by Western allies to high-
from their environments; communicate with their oper- light shared challenges related to embracing software-
ators and other systems; control sensors and weapons defined defence.
systems to achieve mission objectives; and protect mili- Lengthy capability-development processes designed
tary personnel and civilians. for hardware-defined platforms in the mid-twentieth
Capability development and force-generation pro- century prevent armed forces from exploiting the bat-
cesses have not yet accounted for the changing rela- tlefield potential of software as a weapon and a critical
tionship between software and hardware in yielding enabler of enhanced manoeuvrability, mass and force
battlefield effectiveness. Armed forces struggle with protection. This paper argues that software-defined
the development, deployment and upgrading of mod- defence requires a new logic in capability develop-
ern defence software, particularly artificial intelligence/ ment and force generation in which software is the
machine learning (AI/ML). For the United States and horizontally scaled defence platform enabling the inte-
its Western allies, ‘hardware has always been king and gration of conventional capabilities, AI/ML algorithms
software largely an afterthought’.1 Flaws in how defence and other digital emerging technologies, and the real-
establishments integrate software into hardware-based time exploitation of very large amounts of data. This
military platforms generate a high rate of defects in means developing military software and hardware in a
the software. Hardware-dependent embedded soft- defence-as-a-platform approach rather than relying on
ware, which is used in most legacy military platforms, arcane prioritisation between the two. Software-defined
is expensive to maintain and difficult to upgrade. defence entails a greater effort on the digital backbone
Challenges in hardware–software integration and the and the software architecture and design of the force to
continuous use of embedded customised software are enable accelerated adoption of software and horizontal
two of the main drivers of costs and schedule overruns scaling of software- and hardware-based capabilities.
in capability-development programmes. While defence Amid intensifying Sino-American strategic com-
software is recognised as critical to most major weapons petition, two dynamics are shaping the technologi-
systems, for decades it has been ‘widely regarded as the cal defence landscape in Europe and North America.
highest-risk element in an acquisition’, even if software Firstly, China’s investment in software-defined defence
remains the smallest of total programme costs.2 appears to be narrowing the West’s military-power
This report seeks to conceptualise software-defined advantage and rapidly eroding US and European mili-
defence as a fundamental architectural, organisational tary technological superiority. Secondly, as the US is
and operational principle of modern force generation accelerating its efforts to adopt advanced technolo-
and military operations, and to explore its key under- gies in defence, particularly AI/ML, a transatlantic gap
pinnings. In doing so, it investigates ongoing efforts in software-defined defence (capability and doctri-
towards defence software and AI/ML development and nal/operational) has emerged. In contrast to France,
uptake in China, France, Germany, the United Kingdom Germany and the UK, the US has a vision of software-
and the US. The research reviews ongoing policy, defined defence and, though it faces huge challenges
funding and procedural trends in relation to defence to implementation, it is making efforts to accelerate

3 The International Institute for Strategic Studies

the development of software-defined capabilities. The particularly as it pertains to AI-based software solu-
transatlantic gap in software-defined defence may tions. This chapter underlines the data, infrastructure
still be bridged and interoperability challenges miti- and procedural bottlenecks that limit the pace, scope
gated in the medium term. However, this requires that and scalability of modern defence software across
Europeans more boldly embrace agile and iterative the five case-study countries. Chapter Three offers an
defence software and AI solutions; accelerate the digi- interrogation of ongoing defence-innovation efforts
talisation of defence; consolidate their defence spend- in the US and China, and outlines the impact of Sino-
ing and investment in defence software; and accelerate American strategic competition on the international
capability-development processes. system. Chapter Four analyses the efforts towards
The report is structured in five chapters. Chapter software-defined defence of three European case-study
One focuses on identifying what software-defined countries – France, Germany and the UK – and assesses
defence consists of and sets out four key underpinnings their effectiveness. The Conclusion outlines the implica-
of the concept. Chapter Two analyses the challenges in tions for European and transatlantic software-defined
developing and deploying modern defence software, defence and the strategic risks of falling behind.

Software-defined Defence: Algorithms at War 4

1. Conceptual Underpinnings of
Software-defined Defence
The existing literature describes software-defined defence other applications running on the aircraft’s onboard
through the lens of software that is adaptable and agile computer. These are not isolated examples. The MQ-9
in both design and use, defined as the ability to be modi- Reaper and RQ-4 Global Hawk uninhabited aerial sys-
fied continuously and easily without sacrificing per- tems (UAS) deploy around 3,500 lines of source code
formance or operational utility and without having to for operational and control software, comparable to
modify underlying hardware. It is also multifunctional, the 3,000–5,000 lines of source code deployed on M1
effectively sharing the same hardware platform, using Abrams tanks and M2/M3 Bradley infantry fighting
virtualisation as the main tool to scale on demand, and vehicles. The Patriot air-defence system also uses thou-
encompassing the ability to deliver several capabilities sands of lines of code to identify, track and intercept
from a single basic design.3 incoming targets.
This paper conceptualises software-defined defence In short, software defines the function, performance
as a broad architectural, organisational and operational and protection parameters of military capabilities
principle of modern defence strategy.4 The concept is alongside the humans who develop and employ them.
underpinned by the following four defining features, all It is key to upgrading legacy capabilities and building
of which increase speed – particularly the speed of devel- next-generation battle networks. And it has the potential
opment and the speed of deployment and employment. to be more rapidly adaptable than hardware through
continuous upgrades without sacrificing performance,
1.1. The changing relationship between utility or readiness. This does not mean military hard-
hardware and software in generating ware ceases to be important. However, it does signal a
military advantage changing relationship between software and hardware
Defence software is expanding exponentially every in delivering military effectiveness and efficiency on
year in terms of the quantity of code and the com- and off the battlefield.
plexity and autonomy of the tasks performed. In the The rate of technological progress in software and
1980s, the F-16 fighter jet used thousands of lines of related hardware components is superior to the rate of
code to actively control flight surfaces without which technological breakthroughs in conventional military
the aircraft was ‘just a $15 million lawn dart’.5 The F-22 hardware in a three- to five-year period. Because of the
Raptor fighter jet, which was introduced in 2005 as the rate of technological renewal in software, it can add new
best-in-class capability, deploys 1,700 source lines of or improved functionality in weeks or months, whereas
code of avionics software, leading senior US Air Force military hardware can only achieve similar impact over
officials to conclude that ‘about the only thing you can decade-long time frames. Recent research shows that
do with an F-22 without software is take a picture of dual-use technologies such as mobile telecommunica-
it’.6, 7 The fifth-generation F-35 Lightning II Joint Strike tions, LCD displays and Lithium-ion rechargeable bat-
Fighter introduced a few years later and upgraded in teries matured in less than a decade.9 On average, this
2022 deploys 24,000 lines of source code of avionics equals one-third to one-fifth of the time other innova-
software.8 That represents an increase by a factor of 14 tions historically needed to mature, even when allowing
in the complexity and size of the deployed software for variation in the complexity of the technology itself.
between the F-22 and the F-35, without counting the Defence digitalisation initiatives such as the Royal
infrastructure code required to allow the fighter jets Navy’s NELSON programme, the US Army’s Project
to communicate and operate together with ground- Convergence and the US Air Force’s Advanced Battle
command stations and other capabilities as well as Management System could generate new software

5 The International Institute for Strategic Studies

updates on a two-week cycle, with the potential to 1.2. Data-centric architectures and forces
reduce this further to daily releases. Though not all The more software deployed in defence capabilities, the
these software releases are necessarily new code, they more defence data it generates. The human capacity to
still maintain or improve functionality of mission sys- process the exponentially growing volume of data col-
tems. In other, more software-intensive areas, such as lected by battlefield sensors has already been reached.
in the electromagnetic spectrum, software release is For example, military sensors collect terabytes of data
already possible daily and often hourly if conditions per second. One CSO French observation satellite pro-
demand it.10 The US Navy’s Task Force 59 demon- duces over 1,000 high-resolution pictures every day, of
strated in 2022 the ability to release software daily and which human analysts can only process 5% at most.14
upgrade hardware to an AI-enhanced mesh network The MQ-1 Predator and MQ-9 Reaper unmanned aerial
within days.11 vehicles (UAV) generated 14 hours of high-definition
By comparison, mid-life upgrades to military hard- video footage in one mission, of which 99% was wasted
ware and software almost always come decades apart, before Project Maven, the US Department of Defense’s
usually performed at 12- to 16-year intervals from ini- (DoD) flagship AI intelligence-analysis capability, was
tial deployment. Delays in the deployment of next- implemented in 2017.15
generation capabilities often result in the extension of To maximise the disruptive impact of software and
the operational life of legacy systems by employing AI/ML algorithms in defence, a strong data fabric and
defence software or retrofitting them with advanced a mature defence data-management system are needed.
digital technologies. For example, many legacy pro- Previous IISS research found that all five countries
grammes, including the Tomahawk land-attack cruise investigated in this paper struggle with maturing their
missile, the Patriot air-defence system, the F-16 Fighting defence data-management systems. The challenges
Falcon and the F/A-18 Hornet fighter jets have had their stem from several aspects. These include:
operational life extended largely due to performance  difficulties in breaking down siloed data struc-
improvements via software upgrades and reconfigu- tures and fostering enterprise-wide, multidomain
rations.12 Many legacy platforms have recently been data-management systems;
retrofitted with new software-defined sensor suites as  separating data from systems and applications;
sensors remain the largest growing area of investment  providing data lakes and shared data for training
for defence establishments in France, Germany, the UK and development of advanced algorithms;
and the US, as well as the EU and NATO.13 In some  ensuring the timely roll-out of enabling infra-
cases, the physical limit of deploying new sensors, dis- structure, such as cloud and edge computing, and
plays and gear on military platforms has been reached. reliable and secure wideband communications
However, there are incipient efforts within the defence and networking;
industry to transition towards more iterative and fre-  enabling the use of software factories to accelerate
quent software upgrades. For example, Saab in Sweden software development;
performs software upgrades on military aircraft every  consolidating data rights;
two years.  redesigning battle networks to prioritise a cloud-
Finally, hardware alone cannot achieve decision- native software-first approach.
advantage at a time when the real-time exploitation
of data and predictive insights promises to define the In a nutshell, in software-defined defence, military
outer edge of military competitiveness. networks are built to accommodate the flexible real-time
Consequently, Western reliance on conventional mili- sharing and exploitation of data across domains rather
tary hardware (i.e., military platforms) which aggregates than data flows having to accommodate network soft-
sensors and effectors to generate mass and military advan- ware and hardware protocols, as is the case in network-
tage alone is untenable in the medium and long term with- centric warfare. Enabling military battle networks not
out a reorientation towards software-defined defence. just to dynamically share information with each other

Software-defined Defence: Algorithms at War 6

but to use algorithms that can fuse and process all for- personnel resorted instead to unsecure personal mobile
mats of available data is a fundamental quality and func- phones to transmit reconnaissance data.18 In addition to
tionality differentiator for software-defined defence. an insufficient digital infrastructure, the persistence of
The ability to share real-time data with different legacy software systems which have not been serviced in
capabilities (and with allies) across different domains decades, and the struggle to replace hardware they are
is one of the key challenges faced by the US in transi- embedded in, further complicates the problem.19
tioning to the Joint All-Domain Command and Control In 2022, Project Convergence demonstrated the abil-
(JADC2) concept. The high degree of autonomy that ity to share real-time data with other services on a lim-
US military services enjoy has already led to the con- ited basis.20 During the Digital Horizon exercise, the US
current development of three competing multi-domain Navy’s Task Force 59 successfully established real-time
command and control (C2) concepts in the air force data exchange between approximately 13 heterogene-
(the Advanced Battle Management System), the army ous surface and subwater systems via an AI-enhanced
(Project Convergence) and navy (Project Overmatch). mesh network comprising several dozen nodes and
However, all three programmes are struggling to iden- cloud-native infrastructure.21 The exercise demonstrated
tify and deliver capabilities at the moment, often as a how cloud-based software-defined capabilities and AI/
result of the inability of US military assets to share data ML algorithms can enhance situational awareness and
with each other. For example, challenges in data shar- military effects at the tactical edge. However, dial-up
ing between the F-35, F-22A and KC-46A Pegasus air- speed and limitations on data sharing at the tactical
craft and with ground C2 centres significantly delayed edge remain a challenge for most DoD operations.22
the implementation of the air force’s Advanced Battle Significant work lies ahead to normalise such data shar-
Management System programme and cost an addi- ing at the national level, let alone at the multinational
tional USD600m by the end of 2022, without any capa- level within the transatlantic alliance. In October 2022,
bility delivered to the end user.16 Eventually the only NATO adopted a ‘Data Exploitation Framework Policy’
feasible short-term solution was to drop the F-22 from which is an important first step towards a data-centric
the programme. upgrade of federated networks within the Alliance.
This is not a singular case as the US has been strug- However, the framework is limited in relation to NATO’s
gling with system interoperability between individual interoperability and operational needs in a software-
weapons platforms and service command, control and defined battlefield.23
communications (C3) systems. It is particularly diffi-
cult to enable legacy and newer equipment to smoothly 1.3. Software as the core of modular
communicate and dynamically share data. Older gen- weapon and network design
erations of military platforms were never designed with In 2021, the UK Chief of Defence Staff stated that ‘soft-
a digital- and software-first hyperconnectivity frame- ware will be as important as hardware in determin-
work in mind, and newer systems cannot accommodate ing what our Armed Forces will be capable of in the
these outdated protocols. The F-22A and F-35 fighter future’.24 In the US and to a much lesser extent among
jets have incompatible data-link protocols and cannot its European allies, defence software has exponentially
share information with each other.17 Similarly, the navy increased every decade since the 1970s, as has the com-
and army C3 systems for ballistic-missile defence can- plexity and capability of software-defined systems. For
not share information with each other, and the Patriot example, the percentage of system functions performed
air-defence system cannot share data with other military by software rose from 8% of the F-4 in 1960 to 45% of the
assets. During a 2019 NATO demonstration, units of the F-16 in 1982 and to 80% of the F-22 in 2000.25
German 9 Armour Demonstration Brigade could not Nowadays software is a critical part of modern capa-
share data with each other due to the technical incom- bility development. The software deployed on the F-35
patibility of IT and C3I systems. Tactical C2 capability fighter jet is not just an added layer to the aircraft: it con-
was not possible above company level and military trols the aircraft’s aerodynamics, flight and navigation,

7 The International Institute for Strategic Studies

fire-control and weapons systems, sensor-data fusion ‘fundamental architectural limitations derived from its
and analytics, engine, early-warning and safety sys- initial hardware and software design constraints’ and
tems, and more. In short, software makes the aircraft replaced them with an open architecture of ‘dynami-
more manoeuvrable, more lethal, and safer to operate cally loadable real-time tactical combat systems client
on the battlefield. applications, all potentially executing simultaneously
By acknowledging that software is an integral part within the software ecosystem’.28 The F-35 software
of the lethality of a weapons system or of the defence upgrades in Technology Refresh 3 (TR-3), part of the
force more broadly, software-defined defence fosters a Block 4 upgrade, envisage the development of similar
new logic for force generation and capability develop- core combat systems.
ment. In this new logic, software considerations drive Such core software needs to be highly modular and
the architectural design of weapons systems and of adaptative as well as platform-agnostic and hardware-
complex military systems-of-systems to ‘turn a bunch independent. Notably, it can be deployed on multiple
of disconnected hardware products into an integrated platforms, alongside but isolated from other software,
whole that can be operated and managed as a single and can collect, share and exploit data from each of
platform’ that centralises direction and decentralises them. Adaptable software-defined systems share the
execution of individual tasks, in pursuit of human- same hardware platform with other software and algo-
defined mission goals, but without the need for manual rithms which run simultaneously and perform dis-
human intervention.26 tinct tasks. Because they are built by design and used
Enabled by software-defined digital infrastructure, with the purpose of upgrading software-defined func-
defence software and the underlying defence data- tionality iteratively and frequently, such an approach
management system become the common horizontally- maximises the return on investment on both common
scaling platform into which military hardware plugs hardware (including legacy platforms) and on reusable
interchangeably to deliver military effects. In software- digital infrastructure.29
defined defence a common virtualised software platform
can be deployed across multiple military capabilities 1.4. A human-defined and human-centric
in the same family and use application programming approach by design
interfaces (APIs) to enable participation in electronic Finally, while recognising the centrality of software in
and often automatic workflows. In the words of Nand defence capabilities, software-defined defence does not
Mulchandani and Lt. General (Retd) John Shanahan: advocate replacing humans as the most valuable assets
‘Just as Apple runs operating systems such as macOS®, and resources in defence. Software-defined weapon
iOS® (for iPhones), and iPadOS®, the DOD should envi- architectures may enable end-to-end electronic work-
sion a day when it has a TankOS®, FighterOS®, and flows which override the need for manual human
ShipOS® – each running individual hardware systems’ controls of individual weapons systems, but humans
that are nevertheless hyperconnected and highly inter- continue to define the requirements and mission goals
operable.27 A defence-as-a-platform approach, which is for the performance and employment of such capa-
integral to software-defined defence, enables individual bilities on and off the battlefield, and human factors
capabilities to be added or eliminated from the network remain a critical consideration for the development of
in real time, and to push software updates and patches advanced algorithms for defence applications. The key
to all network components and sub-component applica- role of human operators in designing the principles of
tions and systems simultaneously and in real time, with- military human–machine teaming and the core consid-
out degrading or endangering the overall functionality, eration given to the human capacity to process machine-
performance and availability of military capabilities. generated information in the design of new fighter-jet
For example, the 2020 upgrade to the US Aegis com- cockpits demonstrate that software-defined defence
bat system included the development of a Common does not mean the human is out of the loop. Software-
Core Combat System software application to overcome defined defence is premised on enhancing human

Software-defined Defence: Algorithms at War 8

effectiveness and protection across the full spectrum of and certification, as well as new frameworks to ensure
defence tasks and therefore needs to encapsulate ethi- their responsible development and use. Likewise, there
cal and legal concerns emerging from this premise. are different requirements for the development and
Human–machine teaming concepts and capabilities deployment of staff and payroll software, predictive
such as the ongoing British Army ‘Human Machine maintenance software, and target acquisition and fire-
Teaming’ project are examples in which architecture control software which need to be accounted for in a
and design decisions drive not just considerations of software-defined defence approach.
the functionality of the resulting capability but also Finally, the inherent logic behind software-defined
technology test, validation, verification and certifica- defence as a human-centric concept is based on transi-
tion approaches, and development and procurement tioning from a world where it takes over 100 operational
decisions.30 France, too, is in the process of concluding a and logistics staff to operate a MQ-9 Reaper UAV to one
multiannual research study on ‘Man Machine-Teaming’ in which one human can control and direct several mili-
linked to the Future Combat Air System (FCAS) pro- tary capabilities simultaneously. Collaborative-combat
gramme and the technologies that enable it.31 and loyal-wingmen concepts in France, the UK and the
Of course, not all defence software is the same. AI/ US build on the deployment of inhabited capabilities at
ML technologies in particular remain brittle and require the core of a battle group comprising dozens or hun-
tailored approaches to maturing, testing, verification dreds of unmanned capabilities.

9 The International Institute for Strategic Studies

2. A World of Innovation Microscalers

Defence establishments are natural vertical hardware NEXUS platform build on agile and DevSecOps models
hyperscalers as they continuously seek to increase to release new software functionality on a continuous
the number and quality of their defence capabilities. basis. The shift towards agile software development
However, they remain microscalers of software-defined remains slower in France and Germany.
defence-innovation solutions as software-defined inno-
vation efforts fail to scale horizontally. 2.1. Late adapters to a fundamentally
The digitalisation of defence has accelerated over different dual-use technology landscape
the past three decades, albeit at different speeds, in Firstly, the structure of the technological landscape has
China, France, Germany, the UK and the US through radically changed over the past three decades. Nowadays,
the exponential increase in the volume and complexity private-sector actors lead the technological progress of
of defence software. However, governments have not dual-use software and software-defined hardware.
adopted competitive business practices for defence soft- The gap between governments and the private sec-
ware development and procurement.32 Defence software tor in the pace and scope of technological progress and
and AI/ML continue to be developed within waterfall in the adoption of competitive practices to develop
and incremental capability-development frameworks critical advanced technologies is measured in decades.35
which are not optimised for software’s rapid progress.33 Whereas big tech companies started to exploit data and
Efforts to move away from misaligned practices are cloud-computing infrastructure in the early 2000s to gen-
under way in all the case-study countries. Nevertheless, erate software-defined hardware such as the iPhone and
they are complicated by the sheer scale of change in Tesla cars, or software-defined services and functionality
organisational culture which is required to embrace such as Netflix, SpaceX and Uber, most defence establish-
agile and iterative software-development practices and ments (and governments more broadly) are still strug-
relinquish the perception of software as risky and an gling to adopt such practices.36 While the tech industry
inherent source of weakness. In 2021–22, over half of has used digital twins for the better part of the last two
the DoD’s non-classified major capability programmes decades, the defence industry started to introduce them
used agile, iterative and DevSecOps models of software only a decade ago and is still trying to scale their use.37
development, although best practices were not fully Governments and traditional defence-industry actors
internalised and consistently applied. Fewer than one in across the world are outspent, outperformed and out-
three of these projects delivered new software-defined innovated by private-sector actors, big and small.
functionality at intervals of less than six months; fewer In 2021, big tech companies like Alphabet and Meta
than one in ten were releasing software every 2–4 weeks, invested between USD25–35bn (EUR23–33bn) each in AI
as recommended by the Defense Science Board; and R&D, the equivalent of 12–21% of their revenue.38 This is
only one in six was using a software factory in the pro- in addition to similar amounts invested to support the
cess.34 The UK has also begun to experiment with agile development and maintenance of state-of-the-art digital
and iterative software development on a smaller scale. and cloud infrastructure. Defence unicorns like Palantir
In October 2022 the Digital Foundry launched a Defence and Anduril invested approximately USD560m in AI/ML
DevSecOps Service (D2B) to enable continuous inte- research and development (R&D) in 2019–20 and could
gration and continuous delivery of software, although leverage rich venture-capital and private-equity markets
this is still struggling to gain traction within the force. to develop new mission systems and defence products.39
Service-level initiatives such as the Royal Navy’s data By comparison, and despite being the top holders of
and applications initiative and the Royal Air Force’s AI/ML patents in the defence sector, in 2021 the leading

Software-defined Defence: Algorithms at War 10

European defence primes collectively spent approxi- the 1990s.’47 While defence primes are great systems inte-
mately EUR18.5bn (USD21bn) on R&D, on average grators, they have not been as successful at horizontally
3–6% of their budgets, of which AI/ML remains only scaling software-defined architectures and designs in the
a fraction.40, 41 American defence primes invest more same way that big tech and other industries have.
in R&D on average than their European counterparts. For cost-structure and other industry reasons, the
For example, Lockheed Martin invested USD1.5bn defence industry lacks strong incentives to move away
(EUR1.37bn) in R&D in 2021, but AI/ML represents only from decades-long incremental capability-development
a small part of its R&D spending.42 models.48 This often happens with the complicity of
While commercial and military applications of soft- defence establishments, reinforcing a vicious cycle in
ware and AI/ML have different requirements, private- which neither the manufacturers nor the clients have
sector companies in the tech sector push new software any incentive to foster more agile development and
releases every two weeks to three months on average. adoption. This also reinforces a pattern of long-term
By contrast, software maintenance and development in and costly upgrades to defence software, rather than
defence capabilities takes over three years, often focuses adopting a software-as-a-service approach of continu-
on correcting problems in the existing code and algo- ous integration and continuous delivery. For example,
rithms rather than adding new functionality, and comes vendor lock-in for mission-system computers for the
with a high price tag. For example, one of the largest AV-8 Harrier, F/A-18 Hornet and EA-18G Growler air-
cost and deadline overruns in the F-35 programme craft led to three software releases in the space of ten
is the cost for the Block 4 Technology Refresh 3. This is years from the sole designer, developer and manufac-
reportedly over USD4.6bn to 2029, of which USD632m turer of the capability. This does not mean the role of
is software- and AI/ML-related.43 Using an incremen- the conventional defence industry is diminished – that
tal software-development approach, three modernisa- is certainly not the case in terms of systems integration.
tion increments for the F-22 Raptor took twelve years to New market dynamics of closer cooperation between
complete rather than the initially estimated five, during defence primes and non-traditional defence start-
which time projected modernisation costs doubled.44 ups and mid-sized companies are emerging with the
Likewise, maintenance and upgrade costs and timelines potential to impact software development and delivery
are not insignificant. Both the F-22A and the F-35 need schedules. Examples include recent Helsing partner-
three weeks’ maintenance for every 300 flight hours, ships with Rheinmetall and MBDA, Thales’ partnership
during which time the platforms are not available for with Atos, Palantir’s partnership with Microsoft, and
missions, amounting to over USD22m and USD13.4m Atos’ partnership with Amazon Web Services.49, 50, 51
respectively in yearly maintenance costs.45
Governments cannot sustain or improve their military 2.2. Defence software and AI/ML investment
technological edge by relying solely on government- Government investment in defence applications of AI/
driven defence-innovation pipelines, or indeed on the ML technologies as well as the digitalisation of defence
traditional defence industry.46 This is particularly the has nominally grown steadily over the past decade.
case in relation to the development and deployment of Most AI/ML funding continues to be allocated in
disruptive defence applications of software and AI/ML, defence R&D budgets, but assessing real governmen-
where traditional defence-industry actors no longer tal investment remains very challenging because of the
define the technological edge, employ the best human opaque nature of national R&D budgets and the lack
talent or access the range of funding required to remain of clear budget lines for software and digital capabil-
commercially competitive. A 2018 DoD report notes: ity within major capability-development programmes.
‘Software development in the commercial world has The data included in this paper assesses government
undergone significant change in the last 15 years, while budgetary pledges towards AI/ML rather than actual
development of software for defense systems has contin- AI/ML-allocated defence funding, as insufficient public
ued to use techniques developed in the 1970s through data on the latter makes a credible estimate challenging.

11 The International Institute for Strategic Studies

This limitation creates challenges in assessing the which most funds go towards DoD enterprise software.
veracity of average annual AI/ML defence expendi- Nevertheless, there is reason to suspect the DoD’s
ture, especially for countries like the UK, France and AI/ML budgets are understated. In 2020, Bloomberg
Germany which have announced multi-annual budg- Government estimated that the total DoD expenditure
ets for this technology. In such cases most AI/ML funds on AI/ML technologies amounted to USD4bn and pro-
are in fact backloaded by virtue of the growing number jected it to increase to USD5.2bn in 2022. A 2020 CSET
and scope of relevant funded initiatives over time, cre- report assessed DoD AI/ML expenditure between
ating the perception of year-on-year budgetary rises in FY2018–FY2020 at USD11.6bn, amounting to USD3.9bn
defence investment. annual expenditure.54 Calculating conservatively, the
Country comparison of AI/ML defence expenditure estimated defence-wide value of DoD projects includ-
is further complicated by the different national taxono- ing AI/ML in FY23 is USD29.07bn, nearly double the
mies of AI/ML and the variable national costs of AI/ML enacted budget for similar projects in FY22.55 However,
innovation in this technology stack. Investment pledges these costs often include AI/ML as well as other
are also affected by the current high inflation rates dependent hardware, network and human-labour costs.
across the five case-study countries. Estimated costs for AI/ML software development alone
The US and China are the largest spenders on defence range from USD410,000 for automatic test systems to
software and AI/ML by a large margin in comparison to USD1.3m for AI/ML-enabled tactical intelligence col-
European countries, as depicted in Table 1. Nominally, lection and processing on the MV-Osprey platform.56
the US defence budget for fiscal year (FY) 2023 includes Moreover, Govini estimates that overall AI/ML and
approximately USD875m for AI/ML and USD3.98bn on autonomy investment across the DoD, military services
software and digital-modernisation pilot programmes and other US defence agencies may be as high as 50bn.57
mandated by Congress.52 AI/ML expenditure represents Between 2017 and 2021, US defence investment in
0.67% of the DoD’s Research, Development, Test and emerging technologies nearly doubled from USD60.7bn
Evaluation (RDT&E) budget, which in FY23 has seen to USD117.2bn. Recent Govini data suggests this expo-
‘the largest increase in any single account within the nential increase was driven by the response to the
defence budget’.53 This includes a USD200m Artificial COVID-19 pandemic: the largest budgetary increase was
Intelligence and Development Fund created by the DoD in biotechnology, consistent with a response to the pan-
in 2021 to improve tactical AI at combatant commands. demic. However, AI/ML investment remained positive
According to the DoD Comptroller’s Office, in FY23 the across all the sub-stacks, even when adjusted for inflation.
US will spend an additional USD1.62bn on AI/ML, rep- The growth in US defence investment was particularly
resenting approximately 3% of the DoD’s USD57.9bn notable in the fields of decision science (USD3.1bn, with
non-classified budget for IT and cyberspace activities, of a 25.9% year-on-year budget increase), natural language

Table 1: National defence spending and pledged AI/ML defence expenditure in China, France, Germany, the UK and the US*
Country 2022 defence spending 2022 defence R&D spending Pledged annual defence R&D spending on
(USDbn, current) (USDbn, current) AI/ML, 2018–22 (USDbn, current)
China** 242.4 N/A e0.3–1.6
France 54.4 6.6 0.1
Germany 53.4 1.7 0.2
UK*** 71.4 2.2 e0.4–0.5
US**** 766.6 114.7 0.8–e2.5
*Data for China’s 2022 defence R&D spending on AI/ML is not available; estimated expenditure reflects 2020 data presented in Ryan Fedasiuk, Jennifer Melot and Ben
Murphy, ‘Harnessed Lightning: How the Chinese Military Is Adopting Artificial Intelligence’, Center for Security and Emerging Technology, October 2021. The defence AI/ML
expenditure of all other case-study countries is from 2022, the latest available defence data. **China’s R&D defence expenditure is not public and cannot be estimated with
confidence. Data on Chinese defence AI/ML spending for 2022 is not available. ***The UK MoD has not made any public pledges for defence AI/ML spending. The figure in
the table is an estimate based on defence R&D AI projects, Defence Digital annual spending and investments from other innovation funds. ****Estimated US expenditure
calculated based on Office of the Under Secretary of Defense (Comptroller), ‘RDT&E Programs (R-1)’, April 2022’.
Note: e = estimated figure.
Source: IISS, 2023

Software-defined Defence: Algorithms at War 12

processing (USD711m; 19.7% year on year) and machine for an intelligent UAV data-access and -management
learning (USD2bn; 15.3% year on year). Data-at-scale platform, with AI/ML contract costs averaging around
and computer vision were the areas with the smallest CNY1.7m (USD251,000).63
year-on-year budget increases. However, both categories In comparison, France, Germany and the UK invest
benefitted from significant investment, at USD40.2 and only a fraction of Sino-American budgets in AI/ML R&D.
USD2.3bn respectively.58 Encouragingly, other support- In 2021, the British government pledged a GBP6.6bn
ive technologies, including data-visualisation interfaces (USD8.66bn) increase in investment in defence R&D,
and advanced communications, also registered moderate amounting on average to GBP1.65bn (USD 2.16bn) addi-
budget increases over this period. tional annual investments between 2019 and 2023.64 In
In 2017, China pledged to invest CNY1trn 2020, the government pledged an additional GBP100m
(USD148.6bn) by 2030 to become a leading coun- (USD122.2m) increase in defence R&D, in addition to
try in AI/ML technologies. This amounts to roughly the GBP800m (USD963.3m) in the Defence Innovation
CNY76.9bn (USD11.4bn) in annual investments across Fund over the period 2018–28.65 Moreover, the govern-
the civilian and military domains. Under Beijing’s ment pledged a 33% increase in the overall R&D budget
Military-Civil Fusion (MCF) policy, efforts are under- from GBP15bn (USD18.3bn) to GBP20bn (USD24.4bn)
way for the People’s Liberation Army (PLA) to better between 2021 and 2024.
and more rapidly access and adopt cutting-edge AI The Ministry of Defence also set aside an additional
technologies from the private sector. In 2018, China’s GBP4.4bn (USD5.38bn) from 2025–26 onwards to exploit
estimated civilian AI/ML R&D investment was a advanced technologies, including AI/ML applications.66
maximum of CNY39bn (USD5.7bn). Beijing does not The UK will spend an estimated GBP24.9bn (USD30.4bn)
publish a defence budget line for AI/ML. However, on digital, cyber and ICT between 2018 and 2028, with
available assessments of its defence AI/ML investments most individual software-as-a-service contracts in 2022
range between approximately CNY1.8bn (USD300m) averaging in the mid-five-to-six figures. AI/ML RDT&E
and CNY19bn (USD2.7bn).59 In addition, other data and procurement contracts average in the low seven-figure
suggests Chinese defence expenditure on AI/ML was range, as evidenced by the cost of navy’s Intelligent Ship
as high as 1.9% of the PLA’s procurement budget in phase 3, the army’s Human Machine Teaming project
2020.60 In 2021, Beijing pledged a 10.6% increase in its phase 1, the marines’ mine-hunting capability autonomy-
defence basic research and a 7% annual increase in integrator system, and the project TALOS AI-enabled base
its defence R&D spending by 2026. Western sources defence system (i.e., GBP2–3.8m/USD2.4–4.58m). Big data
assess that China’s defence AI/ML expenditure is in analytics contracts, such as the MoD’s SOCIETAS project
the low billions and ‘on the same order of magnitude focusing on electronic warfare mission data and enterprise
as [the] US’.61 business software contracts, are generally more expen-
While China’s resources for AI/ML are very large, so sive (for example, GBP98–150m/USD118–162m) in 2022.67
is the cost of innovation. Inefficiencies in the coordina- The MoD’s estimated digital, cyber and ICT expenditure
tion of military–civil fusion between local and central is GBP24.9bn (USD29.5bn) from 2018–28. The Ministry of
authorities raise the cost of innovation, causing delays Defence’s estimated annual spend on digital in 2021–22
in achieving overall PLA defence-digitalisation goals.62 rose to GBP4.4bn (USD5.3bn), of which 47% was consumed
These internal weaknesses are often obscured, but they by the upgrade and replacement of three core enterprise
partially offset the perceived advantage the Chinese digital systems.68 Cost and deadline overruns in all key
regime has in mobilising and directing resources or digital-capability projects were significant. For example,
accessing private-sector technologies for defence appli- the estimated cost overrun for the Next Generation Core
cation and exploitation. The PLA’s non-classified AI/ Network programme is GBP600m (USD733.2m).69
ML procurement contract costs in 2020 ranged from In 2021–22 the UK invested GBP1.83bn (USD2.24bn)
CNY9,000 (USD1,330) for an intelligent sound-and- in defence R&D, marking a GBP800m (USD939.7m) year-
light alarm detection system) to CNY21m (USD3.1m) on-year increase from 2020, which supports our finding

13 The International Institute for Strategic Studies

that these budgets are backloaded.70 However, budget steadily declining, despite a policy prioritising the
allocations and actual annual appropriations continue to uptake of advanced technologies.77
pose challenges. For example, Defence Digital’s budget The rising costs of ongoing large-ticket projects like
of GBP1.84bn (USD2.26bn) over ten years included a ARTEMIS and SCORPION (Synergie du contact renforcée
GBP658m (USD804m) allocation in the first four years. par la polyvalence et l’infovalorisation) support the assump-
However, only GBP410m (USD501m) was available at tion of growing French defence AI/ML spending. So far
the launch of the digital-transformation agenda, leading ARTEMIS’s budget was EUR6m (USD6.5m) in phase one
to postponement in key capability projects.71 Moreover, and EUR50m (USD54.4m) in phase two, with an estimated
official ministry sources forecast that digital-defence EUR300m (USD327m) under discussion for the newly
expenses would be GBP1.4–3.7bn (USD1.71–4.5bn) over launched phase three.78 In October 2022, the Ministry
the coming decade.72 of the Armed Forces awarded a contract as part of pro-
Comparable investment was announced in France ject TORNADE (Traitement Optique et Radar par Neurones
in 2018. President Macron pledged a EUR1.5bn Artificiels via Détecteur), intended to provide the armed
(USD1.63bn) investment in AI/ML between 2018 and forces with an AI-enabled analysis of electromagnetic-
2022, notably an average EUR300m (USD327m) annual spectrum data, to French company Preligens (formerly
investment split between defence and civilian R&D.73 Earthcube) for EUR240m (USD259m).79
In 2021, France spent EUR6.6bn (USD7.2bn) on defence The Ministry of the Armed Forces announced an
R&D, with EUR901m (USD982m) allocated to support EUR8bn (USD8.63bn) R&D budget in 2023, which
actions for innovation and emerging technologies. Of includes EUR1bn (USD1.08bn) in innovation fund-
the allocated funds, only EUR884m (USD963m) was ing. In addition, spending on defence software and
actually spent on specific projects. In 2022 the ministry AI/ML could also be supported from the ministry’s
underspend on innovation and emerging technologies EUR702m (USD757.7m) space funding, its EUR467m
was even larger, with over EUR100m (USD109m) going (USD504.1m) funding for information-warfare sys-
unused.74 These allocations indicate a trend towards tems, and its EUR5bn (USD5.40bn) maintenance
incremental increases in French investment in techno- budget.80 Since 2018, the Defence Innovation Agency
logical innovation in relation to data, AI and other tech- has been managing a portfolio of 454 projects with
nologies despite limitations in institutional capacity to committed investments of EUR1.4bn, though there is
rapidly absorb increased funding. no public data suggesting that any of these projects
However, the share of these budgets represented by have successfully transitioned to the armed forces at
defence investments in AI is relatively small, with most the time of writing.81 There is also no publicly avail-
funding going to large programmes of record. The able data to ascertain the value of AI investments as
2019–25 military spending plan (loi de la programma- part of the ministry’s much larger procurement budg-
tion militaire) allocated over EUR705m (USD768m) for ets. For example, it is unclear what share of project
AI/ML defence R&D/research and technology (R&T), SCORPION’s EUR10bn (USD10.9bn) budget is dedi-
amounting to over EUR100m (USD109m) annually.75 cated to software and AI development.
However, there is emerging evidence to suggest the Unlike Germany and the UK, which rely on a mul-
budgets are backloaded. In 2020, the French Ministry tinational technological defence base and are often
of the Armed Forces spent EUR61m (USD66.4m) on AI/ happy to procure off-the-shelf tailored solutions,
ML applications through its R&T budget, represent- France’s relationship with its national industrial base is
ing 5.8% of the overall R&T budget, of which EUR21m far more organic. The defence ministry’s 2017 and 2020
(34.4%) was spent on the ARTEMIS (Architecture de IT, cloud and data-management direct-award contracts
traitement et d’exploitation massive de l’information multi- with Microsoft were internally contested because of a
sources et d’Intelligence artificielle) project discussed perceived loss of data and digital autonomy. To main-
later in this paper.76 Furthermore, official data suggests tain analytical and decision-making autonomy, Paris
funding for dual-use technologies in defence R&D is has often taken steps to ensure national overwhip over

Software-defined Defence: Algorithms at War 14

key advanced software providers. For example, in 2020, hardware on thousands of German Bundeswehr inven-
when In-Q-Tel showed interest in buying the French tech tories.86 The same applies to the modernisation of the
company Preligens (formerly Earthcube), two French- Bundeswehr’s Digital Land-based Operations (D-LBO)
based defence private-equity funds, Definvest and Ace and Tactical Edge Networking (TEN) systems, as well
Tikehau, helped raise over EUR20m (USD21.6m) to as to the more recently launched Mission Enabling
keep the company’s French ownership.82 Service Bundeswehr (MESBw), which seeks to provide
Finally, Germany is the only case-study country modular and modern IT systems services and modern-
to spend less than 2% of its GDP on defence.83 Berlin ised C2 systems for all domains for both stationary and
pledged EUR5bn (USD5.44bn) between 2020–25 for deployed forces.87
the implementation of its national AI strategy, which In 2022, following Russia’s invasion of Ukraine,
notoriously excludes defence. In addition, Berlin German Chancellor Olaf Schultz announced that in
reportedly allocated EUR2bn (USD2.18bn) to defence addition to increasing defence spending to 2% of GDP,
AI/ML R&D between 2019 and 2029, amounting to Berlin would invest a further EUR100bn (USD109bn) in
an average EUR200m (USD218m) annual investment. defence over the next decade as part of a special one-off
However, the lack of transparency around the struc- Defence Fund. This included a EUR21bn (USD22.8bn)
ture of the Federal Ministry of Defence’s R&D budget investment plan for digitalisation of defence, including
makes it difficult to ascertain the exact amounts allo- the modernisation of telecommunications, IT and digital
cated and spent on digitalisation and AI/ML respec- infrastructure across the German armed forces. The min-
tively. For example, the 2021 R&D budget amounted istry reportedly only planned to invest approximately
to EUR1.55bn (USD1.69bn), within which there is a EUR421m (USD463m) in defence applications of AI/ML.88
generic R&D/T budget of over EUR1bn (USD1.09bn); France, the UK and the US have set up several
AI/ML applications and defence software are only two government-backed defence-investment funds to
of the technologies which could be funded from this.84 encourage and support the development and matu-
In 2022, Germany’s defence R&D budget amounted ration of emerging technologies, including AI/ML
to EUR1.6bn (USD1.74bn), while in 2023 the budget is in defence applications. Recent examples of DoD-
expected to rise to EUR1.74bn (USD1.9bn), representing backed venture-capital funds are the air force’s
4.51% of total German defence spending, with approxi- AFVentures, a division of its in-house innovation
mately EUR1bn (USD1.09bn) for other R&T activities, platform AFWERX, which invested over USD710m
including digitalisation and AI/ML. Discouragingly, the in new technologies between 2018 and 2020.89 In
funding for key emerging technologies has decreased the UK, the Ministry of Defence established the
steadily from EUR40m (USD43.6m) in 2021 to an esti- Royal Navy’s Defence Transformation Fund with a
mated EUR24.65m (USD26.9m) in 2023, suggesting the GBP75m (USD91.6m) budget; the National Security
federal government’s investment in AI/ML and other Strategic Investment Fund, which comprises seven
digital technologies may also be under pressure.85 independent venture-capital funds, each with invest-
Due to the opacity of R&D budgets, it is difficult to ment of over GBP10m (USD12.2m); and the army’s
accurately assess the ministry’s annual spend on AI or Transformation Fund, with an initial investment of
to establish the share of procurement budgets dedicated approximately GBP66m (USD80.7m).
to digitalisation and AI. The Cyber Innovation Hub’s Similarly, in 2017 France launched the Definvest
EUR10m (USD10.8m) annual budget funds 70 pro- venture-capital fund operated by the French investment
jects (including overhead costs), averaging at just over bank Bpifrance. The fund benefitted from a EUR50m
EUR140,000 (USD151,000) per project. The cost struc- (USD54.5m) initial investment and the ambition is to
ture for the project to retrofit land vehicles with a newly double this amount in five years. By the end of 2020,
designed digital C2 system is estimated to be worth several Definvest had invested over EUR18m (USD19.6m) in
billion euros, with no clarity around the cost of software nine French tech start-ups, and by 2022 official French
development versus the cost of upgrading proprietary sources claimed the fund had doubled its investments

15 The International Institute for Strategic Studies

to EUR100m (USD109m).90, 91 In 2021, the Ministry of the development by virtue of new software-dedicated acquisi-
Armed Forces launched another investment fund, the tion pathways. In the US, nearly half of major capability
DefInnov fund, with an initial investment of EUR200m programmes originating in the army, air force and navy
(USD218m).92 From 2020–22 the Definvest fund under- currently incorporate at least one agile or iterative model
took deals valued between EUR500,000 (USD545,000) for software development – albeit they often do so within
and EUR5m (USD54m).93 the framework of waterfall capability-development mod-
Such award levels are significantly larger than the els.98 Moreover, enabling data and digital infrastructure is
grants and awards that British, French and German often a cascading rather than a parallel process, delaying
defence agencies offer for R&D competitions in AI/ the software’s impact through rapid exploitation by the
ML and more aligned with those in the US, which end user.
range between USD275,000 and USD10.4m.94 In France, For example, Boeing segmented software develop-
Germany and the UK – as well as within the EU and ment for the US KC-46A Tanker programme into small
NATO – such grants and awards generally range iterative development increments which are part of the
between EUR50,000 (USD54,500) in the first round of programme’s overall waterfall development model.99 As
competition and EUR150–250,000 (USD163–275,000) in such, it accelerated the upgrade- and software-release
subsequent rounds.95 Meanwhile, UKRI is Europe’s larg- timetable considerably in comparison to a traditional
est investor in emerging and disruptive technologies, waterfall model. By contrast, according to the US
including AI/ML for defence and civilian applications, Governmental Accountability Office, the challenges
followed by a margin by the European Commission.96 encountered by the F-35 programme were largely to do
with the late development and testing (or lack of test-
2.3. Not agile enough ing) of the software suites.
Scholars and defence analysts in the US and Europe The findings from the US programmes are consist-
agree that software development and acquisition remain ent with capability-development models used in the
a huge challenge for defence establishments in the US UK and France, but not Germany. Indeed, when it is
and, to an even greater degree, in Europe.97 Outdated not procuring capabilities off-the-shelf as the F-35
procurement practices are generally considered one of fighter jet and adding a particular configuration to meet
the main obstacles to software-driven defence innova- German capability requirements, Germany remains the
tion, but they are not the only one. only European country to preponderantly use waterfall
This report interrogates major conventional weapons capability-development models for software develop-
capability-development projects from France, Germany, ment. New structures like the Cyber Innovation Hub
the UK and the US between 2020 and 2022 with a view have developed AI/ML-enabled travel and health appli-
to assessing (a) what capability-development models are cations for the armed forces in six months or less.100
primarily used to develop and adopt defence software; (b) However, lengthy German defence procurement and
how defence software is integrated into complex multian- certification processes would likely preclude warfight-
nual capability-development programmes; (c) what pro- ing applications from being fielded as rapidly.
portion of funding is allocated to defence software and This is partly an organisational-culture problem
AI/ML systems within complex capability-development related to the inertia of entrenched models and proce-
programmes; and (d) how defence software is generally dures. However, it also stems from a leadership-culture
rated in terms of risk to the prospective project. problem in these countries. Agile development and
Firstly, traditional waterfall models remain the principal DevSecOps require a close interaction between develop-
capability-development model for the development and ers, end users and procurement stakeholders in which
adoption of defence software and AI/ML across France, end users iteratively assess and adjust the functionality
Germany, the UK and the US. However, in the US and of tested software to meet their needs and requirements.
the UK, new agile and iterative software-development However, political and military leadership remain
models are beginning to be incorporated in capability fundamentally uncomfortable with this procedure,

Software-defined Defence: Algorithms at War 16

whereby soldiers are actively involved in deciding the resulting in reduced operational capabilities by com-
functionality of future or upgraded legacy capabilities. parison to the legacy Block 30/35 aircraft deployed
Secondly, defence software and AI/ML solutions two decades ago. Like the F-35, the Franco-German
are generally integrated in the later stages of capability FCAS programme, which was reconfirmed in late
projects as customised or embedded software which is 2022, is currently developing a suite of AI-enabled
frequently linked to bespoke hardware.101 Customised situational-awareness, pilot-health-monitoring, and
software is code that is developed specifically for a set real-time data-fusion and -analysis systems. However,
of military requirements; it is platform and hardware it is unclear whether such code will be built on an open
dependent, in opposition to adapted software or com- platform, given the challenges around intellectual
mercial software which can run on commercial and property inherent in the multinational negotiations
modified military hardware. Modified software is com- over the programme.
mercially available code tailored to the needs of the Thirdly, assessing the costs of software and AI/ML
military end user and is often platform dependent, within larger capability-development projects is challeng-
whereas commercial software is purchased off-the-shelf ing. Defence software is frequently the smallest part of a
and directly deployed on license or otherwise. capability project budget despite having a huge impact
Because of proprietary limitations in both soft- on the functionality and performance of the overall capa-
ware and hardware, such defence software is difficult bility throughout its life cycle. The weighted average of
and very costly to upgrade, leading to longer gaps software costs within broader capability-development
between software releases and high costs for mainte- projects in the US is 20.7% for 2021–22, while most projects
nance and upgrade. For example, in the US as well as include a 10–20% software cost share. However, in major
among European countries, the integration of hard- conventional military platforms such as aircraft, ship or
ware and software is still perceived as the highest helicopter programmes, software stands at only 1–2% of
risk when integrating software components in major overall programme costs. For example, defence software
defence capability-development projects. Therefore, within the US Army’s armoured multi-purpose vehicle
instead of 6- to 12-month software release cycles, capa- (AMPV) programme accounts for only 2% of total costs,
bilities incorporating customised software and hard- and 90% of those are customised software.103 In 2021–22,
ware suites often need to wait for mid-life upgrades only 3.8% of 79 major defence-capability projects investi-
10–15 years after they were operationally deployed. gated by the US Government Accountability Office (GAO)
Nevertheless, customised software, including AI had a software component of 80–100% of project costs. In
algorithms, represent most of the defence software 81% of capability projects, software accounted for 20% of
under development in major conventional capability- project costs or less, and in 43% of projects it accounted
development programmes in France, Germany, the for 10% of project costs or less. Nearly half of the projects
UK and the US. Key metrics for budgeting and assess- investigated included 100% customised software require-
ment of such customised defence software revolve ments. Most of the programmes reviewed included soft-
around measuring source lines of code. The latter fits ware costs of USD51–670m. However, software-intensive
a vertical scaling model that remains prevalent within programmes such as the Next Generation Operational
defence establishments rather than a horizontal scal- Control System might spend between USD6.9bn on soft-
ing logic for defence software. ware systems, though it is unclear how much is repre-
For example, the airborne warning and control sys- sented by AI/ML costs.104 The programme plans software
tem (AWACS) aircraft is being upgraded to migrate releases at intervals of 13 months or longer. Of the soft-
‘the hardware and software architectures and applica- ware used, the programme will deploy 37% commercially
tions on the E-3 AWACS aircraft from legacy propri- available software, with 42% being customised software
etary systems to new open architecture hardware and and the remaining 21% modified commercial software.105
software’.102 The use of hardware-dependent custom- In FY22, the number of rapid-prototyping software-
ised software limited the remit of the upgrade itself, intensive programmes in which software accounts

17 The International Institute for Strategic Studies

for 40–80% of the programme costs has increased in Figure 1: Software and AI/ML as a share of major DoD
comparison to FY20 and FY21. This is a testament to capability-development project costs, and average
timelines for software releases (2021–22)
the incremental adoption of agile, iterative or mixed
capability-development models by the DoD and the
Defence software and AI/ML costs as % of major
military services. However, much work remains to be DoD capability-development projects (2021–22)
50%
done. The share of customised software remains high
even across agile development practices – around a
40%
minimum of 40–50% of the total software developed.

% of projects assessed
Moreover, software release times remain on average 30%

longer than the Defense Science Board-recommended

20%
3–6 months or less. The trend towards an acceleration
of defence-software releases in capability development
10%
is not necessarily clear-cut. US official defence data
reveals only 25% of all capability-development projects 0%
0–10% 11–20% 21–40% 41–60% 61–80% 81–100%
release new software in three months or less. The same
Software and AI/ML as % of
data reveals that the same proportion of projects still capability-development project costs

take 13 months or more to release software.106

Reported frequency of software releases in
Finally, defence software is consistently perceived as DoD capability-development programmes using agile
or iterative development models in 2021*
one of the biggest critical risks to capability projects. In the 20%
US, a recent official estimate suggests over 60% of capabil-
% of programmes assessed

ity-development programmes classify the risk associated Government-preferred timeline for

15% new software releases to end users
with defence software to be moderate or critical.107 This is
partly explained by failures to design software at the begin-
10%
ning of the capability design and to iteratively develop and
test software suites. It is also explained by consistent fail-
5%
ures to adequately assess how critical software-defined
capabilities will be to the functioning of the capability
0%
itself, particularly for more hardware-intensive conven- <1m 1–3m 4–6m 7–9m 10–12m >13m

tional programmes, to adequately assess the maturity level Frequency of software releases (months)

of desired software technologies and whether they can be

©IISS
matured in time for delivery timelines.108
In short, the overarching logic of software-defined *32.5% of the projects assessed are not shown because they did not use agile or
iterative software-development models, or had incomplete software data, or did
defence across all case-study countries analysed remains not publish a timeline for software releases.
Source: IISS 2023, based on data from United States Government Accountability
tied to vertical rather than horizontal scaling. As a result,
Office, 'Weapon Systems Annual Assessment', 2021 and 2022.
defence establishments in France, Germany, the UK and
the US, and to a lesser extent China, are proliferating vertically at component or service level. However, they
new initiatives, structures, agencies and programmes in are challenged in scaling horizontally across compo-
the development of modern defence software and AI/ nents, services and agencies.
ML. Examples include the Joint Artificial Intelligence The implication of this trend is that European, and
Center (JAIC) and AFWERX in the US, the Royal Navy’s to a lesser extent American, software capability devel-
former NELSON programme, the Royal Air Force’s opment remains stuck in a decades-old, costly waterfall
NEXUS platform and the French Agence de l’innovation model. The latter has been optimised for vertical scaling
de défense (AID). However, the current life cycle of these of hardware-defined capabilities, but it remains inap-
structures suggests they perform well in developing ini- propriate and inefficient for the horizontal scaling of
tial capability and iterating their innovation solutions software and algorithms.

Software-defined Defence: Algorithms at War 18

3. Software-defined Defence and
Sino-American Strategic Competition
Sino-American strategic competition is the defining US military and bolster national security and economic
feature of the contemporary international system. competitiveness more broadly through the identifica-
Technological competition in AI/ML sits firmly at the tion, adoption and integration of critical and emerging
core of this strategic competition, shaping alliance technologies. A preponderance of these technologies
structures and choices in defence and broader geo- – including cyber, autonomous systems, networked
economic concerns. communications, and augmented and virtual reality
Western competitiveness in software-defined defence – are software-based, while others – such as biotech-
vis-à-vis China maintains a lead for the moment, though nology, hypersonics, quantum sciences and microelec-
that margin is rapidly narrowing as China makes pro- tronics – will require specific software capabilities such
gress in the technological development of AI/ML and as AI to mature.
other digital advanced technologies, supported by its In the military context, the DoD is animated not
growing science-and-technology and industrial base only by incubating and maturing these technologies.
and by access to huge private and public funding. The As articulated in the 2018 National Defense Strategy,
following is a comparative analysis of American and department leaders are aware that ultimate success ‘no
Chinese approaches to software-defined defence. longer goes to the country that develops a new technol-
ogy first, but rather to the one that better integrates it
3.1. US and adapts its way of fighting’.109 To that end, the DoD
Over the last decade, two prominent phenomena have is developing new operational concepts that are predi-
accelerated the DoD’s recognition of the potential of cated on software, software-defined hardware, and
software-based technologies and the need to develop software enhancements to legacy hardware platforms.
the right mix of legacy hardware systems and advanced Perhaps the signature operational concept that focuses
general-purpose technologies such as AI software. the attention of Pentagon leadership is JADC2, which is
The first was the emergence of low-cost, high- an architecture that conceives of every existing or future
performance cloud computing, combined with advances military asset in space, in the air, on land or at sea as
in chip design and processing speed, battery density a sensor. Once connected, this network of sensors can
and supporting materials development, which together seamlessly communicate data; enable AI and machine-
facilitated the ‘internet of things’ revolution. The num- learning tools to generate inferences to speed up deci-
ber of internet-connected devices increased exponen- sions; allow for more precise and efficient actions; and
tially, and the volume of data produced unleashed new ultimately create a safer and more secure operating
capabilities across every economic sector. environment for US warfighters.
The second was China’s evolution into a geostrate- Despite the strategic alignment, however, the depart-
gic competitor with ambitions to reshape the interna- ment’s progress in modernising the joint force to
tional order. A key concern included China’s pursuit of embrace the changing technological environment has
technological leadership in critical and emerging tech- been uneven. The best barometer of the department’s
nologies through comprehensive industrial policies and progress is perhaps its integration of AI software, given
military–civil fusion, accompanied by a range of other that AI, as a general-purpose technology, will drive
licit and illicit technology-transfer strategies. modernisation across existing and future weapons plat-
Against this backdrop, US national-security officials forms, greatly enhance intelligence, surveillance and
in the executive branch and lawmakers in Congress are reconnaissance (ISR) capabilities, and vastly economise
aligning around discrete strategies to modernise the DoD business processes such as logistics, healthcare

19 The International Institute for Strategic Studies

and workforce productivity. Regarding AI uptake in diverse set of analytic capabilities’. Further, the DoD
business and weapons systems, there have been nota- ‘must shape this emerging military-technical compe-
ble successes in delivering outsized impact. However, tition in AI to our advantage while ensuring a strong
owing to long-standing institutional and procedural commitment to military ethics and AI safety. Decisive
path dependencies, organisational and cultural hurdles, warfighting advantage will go to those who integrate
and a general technological immaturity, the department and adapt leading-edge technology to create innovative
still has a long way to go in delivering on these strate- operational concepts with speed and agility.’112
gies at scale. The DoD’s 2021 Software Modernization Strategy
is unambiguous about the role software will play in
3.1.1. Strategic alignment all domains of conflict, underscoring the need for the
A consistent theme in DoD strategy documents is that department to modernise and reform existing processes
emerging technologies are changing the character of to enable decision-advantage through transparent
warfare. The 2022 National Security Strategy, in prior- access to critical data and capability. The strategy puts a
itising ‘integrated deterrence’, seeks to achieve a ‘seam- premium on speed, agile and iterative software develop-
less combination of military (land, air, maritime, cyber, ment, collection and sharing of data, and utilising open
and space) and non-military (economic, technological, architectures to enable constant capability evolution.
and information) capabilities’. Key to this ambition is Recognising that datasets for AI training and algo-
the commitment to modernise the joint force in large rithmic models will increasingly become the DoD’s most
part by ‘investing in a range of advanced technologies valuable digital asset, the Data Strategy underscores
including applications in the cyber and space domains, how important it is for the department to become a data-
missile defeat capabilities, trusted artificial intelligence, centric organisation.113 The strategy highlights that mis-
and quantum systems, while deploying new capabili- sion commanders, warfighters and decision-makers do
ties to the battlefield in a timely manner’.110 not currently have real-time, trusted and secure access
Flowing from the National Security Strategy, the to enterprise-wide data. The strategy lays out guiding
2022 National Defense Strategy outlines the importance principles for breaking down data silos, standardising
of digital technologies in supporting integrated deter- data collection, and encouraging data interoperability
rence: ‘[b]ecause Joint Force operations increasingly and open architectures for development.
rely on data-driven technologies and integration of
diverse data sources, the Department will implement Architectural reforms and budgetary support
institutional reforms that integrate our data, software, for AI
and artificial intelligence efforts and speed their deliv- In a signal to the bureaucracy and to relevant stakehold-
ery to the warfighter’.111 ers in the technology community of its commitment to
This high-level strategy guidance is further real- harness AI solutions for the Joint Force, the DoD has
ised through three critical defence-technology strate- consolidated efforts such as the Joint AI Center (JAIC),
gies: the Digital Modernization Strategy, the Software the Defense Digital Service (DDS) and the Chief Data
Modernization Strategy and the Data Strategy. Officer in a newly created Chief Digital and Artificial
The Digital Modernization Strategy connects the Intelligence Office (CDAO). The role of the CDAO is to:
guidance in the National Security Strategy and National
Defense Strategy through the lens of cloud, artificial lead and oversee DoD’s strategy development
intelligence, C3 and cyber security. Specifically, the and policy formulation for data, analytics, and AI;
strategy highlights that ‘cloud and cognitive computing work to break down barriers to data and AI adop-
will significantly alter warfighting and defense busi- tion within appropriate DoD institutional pro-
ness operations’ and that ‘modernization of warfighter cesses; and create enabling digital infrastructure
support systems will enable improved C2, informa- and services that support Components’ develop-
tion sharing, and decision support, through a rich and ment and deployment of data, analytics, AI, and

Software-defined Defence: Algorithms at War 20

 digital-enabled solutions. Moreover, CDAO will UAS. The counter-UAS technology is deployed at
selectively scale proven digital and Al-enabled DoD infrastructure, including bases, around the
solutions for enterprise and joint-use cases as well world to protect against adversarial drone attacks.
as surge digital services for rapid response to cri-  Synthetic aperture radars (SAR): The DoD is uti-
ses and emergent challenges.114 lising satellite-imagery providers that provide
faster, more capable and higher-quality satellite
3.1.2. Investment support images, day or night, in all-weather conditions.
With respect to AI, the department has made consid- The SAR offering includes machine-learning mod-
erable investments across the military services and the els to augment DoD and US government systems
Office of the Secretary of Defense (OSD) components. to identify relevant objects in troves of images.
Relevant investments in AI at the level of research,  Target recognition: The army is actively develop-
development, testing and evaluation (RDT&E) and pro- ing a target-recognition AI capability to support
curement include: airborne combat operations. The army is also
 From FY2016 to FY2025, the department plans to developing a similar capability to sense and iden-
spend USD14bn on AI.115 tify targets using space-based capabilities such as
 The Department of the Army requested satellite imagery.
USD480.2m for AI in its FY21–FY25 budget, up  Autonomous teaming of AI systems: DARPA
from USD153m in FY16–FY20.116 is cultivating new approaches for the autono-
 In the Department of the Air Force, the AI spend in mous teaming of various AI systems – such as
2019 rose to USD182m, up from USD119m in 2018.117 AI-enabled drones, robots or satellites – that can
 The DoD launched a USD200m Artificial react to new or unexpected situations without
Intelligence and Development Fund aimed at access to centralised communication and human
improving tactical AI at combatant commands control. This is particularly important in contested
by better understanding combatant command or degraded communications environments.120
data and updating their network infrastructure to
improve warfighting capabilities.118 3.1.4. Impediments to AI integration at scale
There are a variety of reasons for the uneven progress
3.1.3. Delivering capabilities in implementing AI within the department and scaling
These investments and organisational manoeuvres are solutions across the joint force. For this analysis, the
delivering novel AI capabilities to many parts of the focus will be on a handful of separate but related issues
military. Through service R&D labs, innovation enti- that have been the most significant obstacles.
ties such as the Defense Innovation Unit, the air force’s
AFWERX and NavalX, and longer-range technology- Organisational complexity
focused entities such as the Defense Advanced Research Firstly, and perhaps most crucially, there is the
Projects Agency (DARPA), the DoD currently has more immensely complicated structure of the DoD. It is one
than 685 AI projects, including some associated with of the world’s largest employers, with 2.8m active-duty,
major weapons systems like the MQ-9 UAV and the joint reservist and civilian personnel. It has over USD3 tril-
light tactical vehicle.119 Additionally, projects include lion in assets under management and conducts major
using AI and machine learning to counter adversarial activities such as acquisitions, command and control,
UAS; enable persistent remote sensing for peacetime global logistics, health and medical care, intelligence,
indications and warning; and enable autonomous team- space operations, facilities management and more.
ing. Some notable case studies are highlighted below. The DoD operates roughly 10,000 operational systems,
 Counter-UAS: The DoD leverages autonomous, thousands of data centres, tens of thousands of servers,
data-fused and AI/ML-enabled sensor technology millions of computers and IT devices, and hundreds of
to detect, identify, track and defeat adversarial thousands of commercial mobile devices.121 Within this

21 The International Institute for Strategic Studies

structure, the reality is that the Pentagon is a collection and responsibilities to strengthen their position
of sovereign entities in the form of the military services, in the competition for resources and to preserve
non-combat agencies and combatant commands that operational ownership in each area. In the context
are in constant competition for resources and are not of pursuing joint concepts of operation, as envi-
necessarily predisposed to overt collaboration. sioned in JADC2, the reluctance to expose data or
Further, the Pentagon is subject to the oversight of allow for API access or common-standards devel-
Congress; lawmakers exert tremendous influence over opment prevents the seamless integration of sen-
strategy by granting authority or approving or deny- sors in different domains. As a result, each service
ing thousands of line items in budgets. In the post-Cold has a derivative initiative – the air force is actively
War era, it can be argued that legislators are motivated developing an Advanced Battle Management
as much (or more) by political-economic concerns as by System, the navy has Project Overmatch, and the
geostrategic ones. army is implementing Project Convergence – and all
are progressing at different speeds and building
Procedural path dependencies with heterogeneous technology stacks.
These structural factors are exacerbated by industrial-  Have vs. need. Over the past half-century, the mili-
age requirements, budgeting and acquisition processes tary, Congress and industry have evolved to resist
that govern the military’s ability to invest in, proto- change and procure more of ‘what the military has’
type and procure technology. Established by Secretary instead of developing ‘what the military needs’
McNamara in the 1960s, what is now the Planning, for the next conflict. Current incentives, through
Programming, Budgeting and Execution (PPBE) pro- the defence-budgeting process in particular, com-
cess was appropriate for an era when the DoD was the mit trillions of dollars to tanks, ships, planes and
exclusive investor in and purchaser of technology (such nuclear weapons at the expense of fielding alter-
as microelectronics) and the desired products were native concepts and capabilities, investing at the
exquisite, hardware-dominant weapons platforms such right level in new domains such as AI, space and
as aircraft carriers or tanks, whose production cycles cyber, or investing to support the industrial base
were more linear and predictable. in new technologies such as small drones or com-
Software, however, is never finished and its develop- mercial satellites.
ment is non-linear. It is constantly undergoing upgrades  Prime economics. The incentives created in the
to improve functionality and to patch vulnerabilities. DoD cascade through the defence-industrial base.
Until very recently, the DoD would shoehorn software Specifically, the prime contractors understand
acquisition into hardware-acquisition pathways, lead- the economic advantages of maintaining the sta-
ing to significant delays, cost overruns, glaring cyber tus quo. In a world where modular open-source
vulnerabilities due to lack of upkeep, and a rapid obso- architectures and easily available source code
lescence as final products were not keeping pace with and data are the norm in the commercial sector,
industry standards. the primes are motivated to deliver vertical solu-
tions where they own the intellectual property of
Mismatched incentives the entire technology stack and are not inclined
Flowing from the impediments listed above, incentive to build open architectures. This contributes to
structures for stakeholders often work at cross purposes to the ineffectiveness of many DoD software pro-
the goal of ensuring that the US military is AI-ready, leav- grammes. According to Eric Schmidt, the chair-
ing the US less prepared to defend itself in the AI era.122 man of the National Security Commission on
 Rice bowls. Each military service retains high Artificial Intelligence, ‘costs spiral due to the
degrees of autonomy over research, develop- need for legacy-system support, while the soft-
ment and acquisition processes. As such, service ware is rife with cyber vulnerabilities. When soft-
leaders tend to guard programmes, resources ware fails, the cost to fix it can run in the tens or

Software-defined Defence: Algorithms at War 22

 hundreds of millions, and even then it may ship underlying these models are not communicated to
with reduced capability.’123 the end user. In a military context, treating some-
thing as truth can have more serious consequences
3.1.5. AI focus going forward than in the commercial sector, so the DoD will
As articulated by the Chief Digital and AI Officer, the work with vendors to improve the design of mod-
DoD recognises that certain practices, such as vendor els so that the context around inferences is com-
delivery of vertical AI solutions, will continue. While not municated to the end users.
ideal, it is the only way that AI can currently be deliv-  Cultural and organisational. Building AI that
ered. Moving forward, the CDAO will focus on aligning is both responsible and functional requires not
efforts across the military services on several discrete only algorithmic and infrastructure development
issues to improve the provision of AI. These include: but also organisational change. The DoD was
 Data readiness. The DoD has exabytes of data not built as a computationally enabled technol-
and potentially more. Today, only a fraction of the ogy company. It is similar to a non-tech company
data is ready to develop and train AI. The CDAO from the Fortune 500 (e.g., a bank) that has core
has articulated a vision where, in the hierarchy competencies in other areas and needs to leverage
of needs, DoD must first get the data layer right. AI to accomplish its goals. The DoD has a hetero-
To that end, the CDAO will work to find the 15% geneous set of computational systems built over
of the data tables that drive 60–70% of the enter- decades, most of its staff are not AI experts, and it
prise value, and aim to centralise those. If the DoD was not built with data as a first-class citizen. This
can get those correct, it can allow AI developers means that, in addition to updating computational
and application developers to write back into the hardware and software systems to rapidly create,
DoD’s data lake.124 update and continuously deploy models at the
 Models as a service. The DoD currently has very speed required, the DoD needs to consider how
few technologists trained in building models. The best to build an AI-driven organisation that does
DoD should facilitate this training and leverage it not force AI into existing manual workflows, but
in order to keep pace with the commercial sector. rather co-optimises human and machine inputs to
 Testing and evaluation as a service. Importantly, best accomplish a given set of end tasks.
the DoD lacks a robust process for evaluating AI
performance as well as the tools to course-correct. 3.2. China
To that end, the Department has drafted a test- Under President Xi Jinping’s rule, China has aimed to
and-evaluation framework specific to AI which become a leader in emerging and disruptive technology
is intended to provide a sequential process for as part of a wider effort to transform itself from a manu-
verifying and validating an AI capability, in line facturing hub to a technological powerhouse. This effort
with the DoD’s ethical principles and focusing has both military and civilian dimensions, whereby the
on ensuring security, resilience and robustness. Chinese Communist Party (CCP) views AI and ML as
Moreover, this should provide independent and central to driving economic growth and national com-
unbiased assessment of the quality and readiness petitiveness, as well as key to military transformation
of AI-enabled systems to increase confidence by and building combat capability for a modern PLA by
end users. 2035 and a world-class military by 2049. Militarily, the
 AI by design. Currently, the DoD utilises AI mod- Chinese government has directed the PLA to become an
els that communicate various inferences, and informatised (信息化) and intelligentised (智能化) mili-
users treat that output as truth. However, underly- tary between 2020 and 2049, which will require increas-
ing these outputs are varying levels of confidence ing the integration of digital and information-based
that could introduce grounds for less confident technologies as well as intelligent technologies into the
interpretations. The certainties and uncertainties military’s organisation and capabilities. To this end,

23 The International Institute for Strategic Studies

the Chinese government has published national-level of normalised communication and coordination mech-
strategies as well as supported national investment in anisms for scientific-research institutes, universities,
both the public and private sectors. Through a national- enterprises and military-industrial units. In terms of
level strategy of MCF (军民融合), the government has application, the plan emphasises a focus on AI technol-
also promoted the building of linkages between the ogy for ‘command decision-making, military deduc-
civilian and military industries that aims to go beyond tions, and national defense equipment’.
the attempts of civil–military integration (军民结合) to Since 2017, national defence white papers and Party
create synergies to such an extent that both industries Congress work reports have continued to emphasise
jointly drive forward national economic and military the importance of emerging and disruptive technolo-
development. However, despite these efforts, China’s gies for PLA joint capability development.127 Firstly, the
approach is not without difficulty and obstacles to suc- 20th Party Congress work report states that the PLA
cess, and while the PLA and its affiliated research insti- will ‘study and gain a good grasp of the characteristics
tutes have become more interested in researching and of informatised and intelligent warfare and the laws
developing AI/ML for military use, integration of these that govern it, provide new military strategic guidance,
technologies remains limited. and develop strategies and tactics for a people’s war’.
Secondly, the document puts forward specific areas for
3.2.1. National AI/ML strategies and relevance improvement. Thirdly, it mandates that the PLA will
for warfighting ‘increase the proportion of new-domain forces with
The Chinese government has published 11 guiding new combat capabilities, speed up the development
documents and white papers related to AI/ML. These of uninhabited, intelligent combat capabilities, and
include the 2015 ‘Made in China 2025’ strategy, and the promote coordinated development and application of
2021 ‘14th Five Year Plan for National Economic and the network information system’. Fourthly, the report
Social Development of the People’s Republic of China states that ‘high-tech training’ will be used to intensify
and Outline of the Vision for 2035’.125 The National New efforts to train the PLA. Lastly, the PLA currently aims
Generation AI Plan (新一代人工智能发展规划), pub- to be able to disrupt an opponent’s use of AI/ML and big
lished in 2017, is most relevant to the development of AI data, in what one AMS researcher calls ‘data warfare’.128
for defence. The plan established initiatives and goals With regard to MCF, the 20th Party Congress work
for R&D, industrialisation, talent development, educa- report states that the government will ‘better coordi-
tion, standards-setting, and regulation of ethical norms nate strategies and plans, align policies and systems,
and security.126 According to the plan, China’s AI indus- and share resources and production factors between the
try should have been ‘in-line’ with competitors by 2020, military and civilian sector’.129
and should reach world-leading status in some AI fields However, challenges were also identified in the 2017
by 2025 and become a primary centre for AI innovation National AI Plan, which implied that China, while having
by 2030. advantages, was starting from a low and incoherent base.
The plan emphasises the importance of MCF to the The document stated that ‘we must be soberly aware that
development of military and civilian AI, and therefore there is still a gap between the overall development level
promotes collaboration between the two sectors at foun- of AI in China and developed countries’. In particular, the
dational levels, such as in data- and platform-sharing. challenges identified were a lack of original achievements
For example, the document states that public-data in basic theory, core algorithms and key equipment, high-
repositories, standard-test datasets, service platforms end chips, major products and systems, and basic materi-
and other sources of information should be provided for als. The document also stated that the country lacked a
AI platform-testing and evaluation – including between comprehensive ecosystem and industrial chain of compo-
military and civilian actors. Following that, the plan nents, software and interfaces, scientific-research institu-
promotes the joint development of basic theory and tions and enterprises, R&D talent, infrastructure, policies
key common-technology R&D, and the establishment and regulations, and standards systems.

Software-defined Defence: Algorithms at War 24

3.2.2. National AI effort The PLA continues to gather and develop method-
According to the 2017 National AI Plan, the Chinese ologies for the use of big data (大数据), to the United
government wanted to cultivate an AI industry worth States’ concern. For example, the FBI determined that
CNY1trn (USD148.6bn) with related industries worth the PLA was linked to the hack of Equifax in 2017.133
CNY10trn (USD1.48trn) by 2030. In Beijing, the govern- Experimentation at lower levels of the military has also
ment has built an AI research technology park worth been reported. In 2021, the commander of the Hebei
USD2.1bn together with Chinese technology champi- Armed Police Corps and member of the National
ons.130 Specific estimates of how much China spends on People’s Congress Rong Jiuhua discussed in an inter-
AI-enabled technology for military purposes are not avail- view with the PLA Daily his own experimentation with
able. Some estimates have been made for China’s total AI big data.134 For example, he reportedly led an R&D
R&D spending. Initial estimates by Ashwin Acharya and team to explore new ways of building and improving
Zachary Arnold suggest with low to moderate confidence the command system of the Hebei Armed Police Corps
that China’s public investment in AI R&D was in the using image maps. Rong also expounded the value of
order of a few billion dollars in 2018 and is significantly integrating big data and blockchain technology to help
lower than previously thought.131 Additionally, the same promote ‘the intelligent development of the PLA and
researchers hypothesised that the total value of a break- continuously improve the joint combat capability based
down of AI R&D spending in 48 plausibly AI-related on the information network system’.135
Chinese government guidance funds may have been At Airshow China 2022 in Zhuhai, uninhabited
between USD2bn and USD8bn in 2018. platforms and systems were on full display. Aviation
As a result of a national and top-down effort, the Industry Corporation of China (AVIC) displayed the
Chinese government was able to address some of the Wing Loong 10 (WL-10) UAV in colour, suggesting that
previously mentioned challenges by 2022, both in R&D, the UAV has already entered into service. Chengdu
application and integration, and governance of AI Aircraft Industry Group (CAIG) unveiled a possible
within the miliary domain. Detailed information about mock-up of the new Wing Loong 3 (WL-3) armed drone.
the central government’s resourcing, subsidisation, and China Aerospace Science and Technology Corporation
efforts to research and develop as well as integrate AI (CASC) displayed a mock-up of the FH-97A UAV, with
and ML into the PLA is mostly unavailable in open- a computer-generated presentation that showed it oper-
source literature. However, some reporting does exist ating in an air-to-air networked role with manned J-20A
that provides inferences into where the PLA might cur- stealth fighters. China Aerospace Science and Industry
rently stand in its achievements. Corporation (CASIC) displayed an example of the Tian
For example, cross-sectoral and national-level initia- Ying (Sky Hawk) fixed-wing drone, which could fulfil
tives continue, such as the annual Data Forum, which a UCAV/ISR (uninhabited combat aerial vehicle/intel-
is sponsored by the Academy of Military Science and ligence, surveillance and reconnaissance) role and pos-
attended by 500 leaders from the Central Military sibly be deployed on aircraft carriers. However, it is
Commission, theatre commands, services and arms, uncertain whether all these autonomous and uninhab-
the Academy of Military Science, the National Defense ited technologies are also described as ‘intelligentised’.
University, and the National University of Defence Land-based uninhabited technology was also on dis-
Technology. The theme of the 2019 conference was play, and older reports suggest the PLA ground forces
‘military big data to promote the development of mili- have converted older capabilities into uninhabited vari-
tary intelligence.132 Published reflections on the confer- eties with some level of suggested intelligentisation.
ence highlight areas of discussion, from the importance The PLA Navy has also been testing uninhabited intel-
of big data, to building network infrastructure within ligent underwater capabilities.136 A paper published by
the PLA to achieve informatisation and share data and the chief designer of China’s J-20A fighter aircraft, Yang
intelligence across the military, to the strategic impor- Wei, also stated that China will leverage AI to help pilots
tance of big data and AI/ML for future warfighting. process information at each step of the OODA loop in

25 The International Institute for Strategic Studies

air combat. Chinese researchers have also experimented the private civilian and defence sectors.140 Secondly,
with the use of AI in cyberspace security and combat.137 Andrea and Mauro Gilli indicate that the limitations of
While the PLA’s capability development is likely to be imitation, reverse engineering and cyber espionage mean
more advanced than suggested in open-source infor- that China’s ability to leverage foreign innovation in areas
mation, it is unlikely that the PLA currently fields fully where it might be lagging behind does not always easily
intelligentised capabilities, which PLA and government extend to the PLA and China’s domestic defence indus-
reporting, such as that in the 20th Party Congress work try.141 Lastly, as Elsa B. Kania points out, China’s need to
report, suggests is still aspirational. import the advanced AI chips necessary for AI/ML ena-
In addition to platforms and systems, the PLA has blement and deployment is a particular weakness.142
reportedly also been using AI to train pilots. According The United States’ export controls on semiconductor
to one report, the PLA Air Force has deployed AI as chips to China have complicated this picture even fur-
simulated opponents in pilot combat training. The PLA ther. The most recent of these controls, announced on 7
Central Theatre Command Air Force held a training October 2022, aims to permanently freeze the Chinese
simulation in mid-2020 during which top pilots faced AI military and civilian high-tech industries where they cur-
in exercises. The report states that ‘the AI technology- rently stand, and to further increase the gap between the
based autonomic aerial combat simulator was jointly US and its allies and China in terms of advanced techno-
developed by the brigade and research institutes’.138 The logical capabilities. The measure prohibits US companies
report also states that data from exercises is looped back from exporting technology, software and equipment to
into the AI simulator for further machine learning. China that is used to produce advanced computing chips
In terms of governance of the use of AI in the military and supercomputers, and prohibits US citizens from
domain, in 2021 the government published China’s first supporting certain China-based chip companies with-
‘Position Paper on Regulating Military Applications of out a licence from the US government.143 The new rules
Artificial Intelligence’, setting forth its position on AI also have an extra-territorial element, as they will impact
security governance.139 other chip-making producers such as ASML Holding
It is important to note that in all cases, R&D and the N.V. in the Netherlands. The ultimate impact of these
application of AI/ML in the military domain remain a regulations will be determined by the ability of the US
work in progress. For example, a researcher at AMS on and other countries to enforce compliance.
big data, Lu Bin, has stated that military big data is still China remains a net semiconductor importer, particu-
poor in quality, low in value density and incomplete larly in advanced chips. In 2021, China imported semicon-
or false. While AI/ML requires sample data, the PLA ductors worth USD432bn, and in 2019 Beijing imported
has very little data on current wars and, logically, no 84.3% of the mature-generation semiconductors it used.144
data on future wars. Furthermore, informationisation, President Xi has made self-sufficiency in semiconduc-
which is required for enhanced use of big data and tors one of his main priorities. The state-backed China
application of AI/ML throughout the military, has still Integrated Circuit Industry Investment Fund is the larg-
not been completed. est of such funds in China, with CNY343bn (USD50.6bn)
set aside for advancing the country’s chip industry. The
3.2.3. Further challenges local governments of Beijing and Shanghai also have their
Previous research has already identified that the PLA’s own semiconductor-industry funds worth CNY32bn
development and integration of AI/ML will face several (USD4.72bn) and CNY50bn (USD7.38bn) respectively.145
challenges. Firstly, according to Tai Ming Cheung, the Furthermore, in 2022 some foreign companies attempted
PLA’s defence acquisition and innovation systems include to develop semiconductor lines of production solely for
the monopoly structure of the Chinese defence industry, the Chinese market, thereby escaping export controls set
bureaucratic fragmentation and weak acquisition mecha- by the US government.
nisms. These structural weaknesses can also be hurdles However, Chinese domestically manufactured
to military–civil fusion and to achieving synergy between semiconductors are not equal in processing capacity

Software-defined Defence: Algorithms at War 26

to those procured internationally, and developing Furthermore, China’s national semiconductor fund has
indigenous advanced semiconductors remains a dif- been undermined by anti-corruption investigations.148
ficult challenge.146 China’s own microprocessor indus- Without the ability to produce advanced semiconduc-
try manufactures significant quantities of chips from tors domestically, China’s capacity to resolve existing
24nm and up, but experts argue that Beijing is still bottlenecks and progress further towards building an
a way from being capable of producing micropro- informationised and intelligentised military will only
cessors at the most advanced production nodes.147 become less realistic.

27 The International Institute for Strategic Studies

4. European Approaches to
Software-defined Defence
As discussed in the previous section, the US and and the FCAS programme. Large-scale projects are pro-
China have a mixed track record of partial successes gressing more slowly in the UK.150 Notable examples
and partial failures in embracing software-defined include the the Tempest fighter jet, the Royal Air Force’s
defence. But what does this entail for Europe? What PYRAMID project, which seeks to establish a reusable
perils could be associated with European powers’ own and open-systems architecture and reusable software
efforts to develop software-defined defence as Sino- components for legacy and future air platforms, in
American strategic competition intensifies? This sec- order to reduce software-development costs; the Royal
tion will analyse the British, French and German efforts Navy’s new Naval Strike Network, which incorporates
to implement software-defined defence and assess the inhabited and uninhabited vessels in battle networks
challenges that stem from Europe lagging behind the commanded by the Maritime Autonomous Platform
two global superpowers. Exploitation (MAPLE) platform; and the British
European approaches to software-defined defence Army’s Human Machine Teaming project, which aims
are more modest in scope and ambition when compared to develop and deploy a human–machine team brigade
to those of China and the US discussed in the previous by 2025. In contrast, the German armed forces remain
section. France, Germany and the UK prioritise national among the least digitalised in Europe, which translates
AI/ML development in a variety of sensitive domains, into a different and more modest set of priorities in
including electronic warfare, ISR and strategic C2, as relation to defence software.
well as in capabilities related to the French and British The three European countries’ efforts towards
nuclear deterrents.149 European countries are often software-defined defence share an emphasis on data
happy to be fast followers of defence-technological pro- and a long-term perspective on greater integration
gress in the US, and defence software has been a tradi- of modern defence software and AI/ML in defence,
tional area of weakness of some European states. which is linked to next-generation capability pro-
French, British and German efforts focus predomi- grammes like the FCAS, the Tempest fighter aircraft
nantly on accelerating the digitalisation of defence and the Main Ground Combat System (MGCS). This
capabilities and put data squarely at the centre. In addi- long-term perspective, mainly as a result of budget-
tion, France and the UK are pursuing the deployment ary limitations, frustrates opportunities to upgrade
of AI/ML algorithms in defence, underpinned by solid legacy systems in the three countries’ inventories
trust and responsible vetting procedures. In Europe, using modern defence software and AI/ML technolo-
only France and the UK have developed and published gies. Because their military capabilities, systems and
defence AI strategies and matured an expansive eco- networks are less digitalised and more fragmented
system to enable defence innovation, technology matu- than American ones, the challenge for European
ration, venture-capital funding opportunities and more approaches to software-defined defence is that they
in support of their efforts. Both France and the UK require a comparatively higher upfront investment
underline the importance of multi-domain operations in developing defence-as-a-platform and adopting
underpinned by human–machine teaming and col- a software-as-a-service approach, even as they con-
laborative combat between inhabited and uninhabited tinue to struggle with insufficient technical talent to
capabilities. However, only France is actively imple- advance in-house defence-software development.
menting large-scale projects to this effect, including However, Europeans also diverge significantly in their
SCORPION and VULCAIN, through which Paris seeks approach to – and concerns around – software-defined
to increase the number of autonomous land vehicles, defence because of differences in their defence m is

Software-defined Defence: Algorithms at War 28

consolidated as the almost exclusive basis for sovereign defence, which is underpinned by key strategic and pol-
advanced capabilities for the French military. Resulting icy documents. The 2021 ‘Integrated Review of Security,
capabilities still encounter important interoperability and Defence, Development and Foreign Policy’ empha-
communications challenges because of proprietary indus- sises the importance of sustaining strategic advan-
trial requirements driven by financial incentives. However, tage through science and technology. The ‘Integrated
the close relationship between the French defence estab- Operating Concept’, first published in 2020 and updated
lishment and its national industrial and technological base in 2021, highlights that ‘technologies such as pervasive
also provides Paris with potential solutions to alleviate this availability of data via enhanced cloud connectivity,
challenge, and to a larger extent than London and Berlin, machine learning and artificial intelligence, and quan-
who struggle to influence international arms producers. tum computing will allow not just a new generation of
weapons systems but an entirely new way of warfare’.154
4.1. UK This has translated into an increased investment in
Within the UK defence establishment, there is a firm cyber and digital capabilities in the ministry’s ‘Defence
belief that future battlefield advantages will no longer Equipment Plan’ for 2022–32.
be determined by superior hardware capabilities but Furthermore, the UK’s Data Strategy for Defence
rather by a combination of people, hardware and soft- emphasises that by maximising value from data via an
ware: personnel and hardware will remain instrumental integrated digital environment across defence, decision-
to warfighting, but software will allow for secure pro- makers can take decisions faster by using software to
cessing, analysis and exploitation of data at the speed of derive insights from data, resulting in warfighters gain-
relevance.151 Software and data are expected to enable ing real-time access to information and, in turn, opera-
decision-advantage for British troops and decision- tional and decision-making advantage.155 And the Digital
makers. The combination of deployed military software Strategy for Defence released in April 2021 underlines
and hardware is expected to regenerate lost mass and the importance of data-driven systems ‘where “software
manoeuvrability for the British armed forces.152 And defined capability” gives (…) an asymmetric edge by
algorithms may even offer asymmetric advantage on sensing, recognising and responding to new opportu-
the battlefield in the event of defending against Russia nities and threats’ faster than adversaries.156 Under the
and China as part of multi-domain integration.153 Digital Strategy for Defence, the UK aims to achieve
three strategic outcomes by 2025: a ‘Digital Backbone’
4.1.1. Data as the source of asymmetric advantage enabling the flexible, seamless and secure real-time shar-
The UK’s approach to software-defined defence puts ing of data across the enterprise; a ‘Digital Foundry’ that
data at the centre of efforts to adopt and integrate soft- is the ministry’s software and data analytics factory; and
ware and emerging technologies in defence applications a skilled community of digital specialists who can sup-
to achieve an asymmetric advantage against strategic port the ministry’s digital transformation.
adversaries like Russia and China. Given the emphasis In 2022, the UK released its Defence Artificial
on data as a core resource in future warfare, the Ministry Intelligence Strategy, becoming the second European
of Defence has adopted a whole-of-defence, system-of- country to publish such a programmatic document. The
systems approach to reorganising and restructuring document adopts a ‘“systems” perspective’ for AI pro-
approaches to capability and software development, curement that focuses on ‘outcomes, delivered through
and views disruptive and emerging technologies as clear frameworks and processes, and guided by [the]
central to force structure, procurement strategies and conviction that AI can be a powerful force for good’.157
its overall approach to future conflict. The document proposes a pathway to enhance exist-
ing safety and regulation practices, in compliance with
4.1.2. Policy alignment democratic principles and international law, and sets a
There is alignment across different defence stakeholders clear framework to ensure UK adoption and use of AI in
around a strong sense of British prioritisation of data in defence is trustworthy and responsible.158

29 The International Institute for Strategic Studies

 However, human factors and digital enablers of AI/ widespread recognition that technology teams across the
ML such as data, software and digital infrastructure defence establishment have not yet fully adopted new
remain important challenges. The strategy does not standards and practices. Defence Digital’s programme-
identify or prioritise a set of software-defined capabili- and project-delivery track record continues to be affected
ties, though it does underline that the UK is seeking to by ‘a lack of skilled and experienced personnel, immature
deploy AI/ML to achieve strategic and operational project controls, and a culture focused on the approvals
advantage against adversaries.159 The document under- process rather than outcomes’, which undermines its
lines a high level of ambition, based on an awareness of ability to affect change across the defence sector in the 90
the fact that London is not competing at the level of either programmes it currently coordinates.161
Washington or Beijing: ‘Our vision is that, in terms of AI, The sense of muddled UK capability priorities is rein-
we will be the world’s most effective, efficient, trusted forced by recent announcements about cancellations of
and influential Defence organisation for our size.’160 In future capabilities such as the Mosquito uninhabited
an effort to manage expectations, the strategy underlines loyal-wingman programme, which seemed to be at the
that in the short and medium term the UK expects to core of regenerating mass for the Royal Air Force and
deploy mainly enterprise-and-support AI rather than enabling collaborative combat alongside inhabited and
operational battlefield algorithms. Further clarity is uninhabited versions of the Tempest fighter aircraft and
expected from the Implementation Plan of the Defence uninhabited Alvina drones.162 Another relevant example
Artificial Intelligence Strategy and from potential ‘AI is the unexpected reform of the Royal Navy’s NELSON
readiness’ guides for individual units on the develop- programme, an internationally appreciated brand for
ment and adoption of AI/ML to de-risk the technologies the service which was increasingly successful in regular
at a more granular organisational level. and frequent software releases.
More recently, UK Strategic Command announced
4.1.3. Software-defined capability priorities a focus on three priorities in software-defined capabil-
The focus in UK capability development remains on ity development in the coming years: cyber, the electro-
how data can enable new digital technologies to be magnetic spectrum and enhanced situational awareness.
exploited and scaled in the armed forces, rather than on However, there is little detail about how software-defined
the actual process of shifting from a hardware-oriented defence will shape the UK Strategic Command’s capabil-
to a data-based and software-defined defence approach. ity priorities without the capability provided by the Digital
The Digital Backbone and Digital Foundry have become Backbone, which has been slower to achieve than initially
the focal points of the ministry’s efforts. Defence planned. In January 2023, the Ministry of Defence pub-
Digital’s slow progress towards building a capability lished ‘Joint Doctrine Note 1/23’, which provides a blue-
that incorporates a common standards-based, interop- print of changes in methodology, practices and capabilities
erable system throughout the ministry has often frus- underlying ISR to accommodate both more data-varied
trated other software-defined defence initiatives within and data-intensive practices as well as to facilitate end-to-
the military services. This common standardised foun- end electronic workflows in ISR.163 However, more work
dation developed by Defence Digital would be expected is needed to flesh out mature doctrine and operational
to allow Defence to exploit curated data as a strategic concepts that build on and fully exploit a software-defined
asset and to enable a strategic shift to software-intensive defence approach enabled by data, software and AI/ML.
capabilities. The UK’s challenges in building a mature
and reliable digital backbone for its armed forces are 4.1.4. Software-defined defence
not unique, as the US DoD’s ongoing struggle to estab- and capability development
lish a Joint Common Foundation has proved. Software-defined capabilities are included as part of
Despite tangible progress in the implementation of efforts in adopting emerging technologies in defence,
the Digital Backbone and Digital Foundry, in line with and there are efforts across the defence sector and
agile models of capability development, there is also the services towards incorporating agile and iterative

Software-defined Defence: Algorithms at War 30

software capability-development models into R&D to deliver an adaptable and robust system based on an open-
and capability-development pathways. The UK system architecture built for multi-domain integration
Strategic Command’s jHub is developing and test- and interoperability with allies. The programme’s official
ing a ‘Sustainable Tech Adoption Model’ (STAM) that description suggests it uses an approach based on the disag-
is exploring new ways of working with industry to gregation of sensors, human deciders and effectors, work-
procure new capabilities.164 Next-generation flagship ing in an end-to-end electronic workflow environment
capability-development programmes like Tempest are where command is centralised and tasking decentralised.
implementing a ‘digital-first’ approach to digitally engi- Similarly, the Royal Navy’s approach to the imple-
neer, build and test the concept. And the UK already mentation of the Digital Strategy for Defence highlights
deploys several defensive systems enabled by rudi- several principles consistent with a software-defined
mentary automation, data-management-as-service and defence approach, including a systems-of-systems
AI-powered analytics.165 approach of disaggregating sensors, human decid-
In 2022, Defence Digital signed a three-year GBP75m ers and effectors that is enabled by an end-to-end
(USD90.3m) contract with Palantir which will help AI-enhanced architecture, has a cloud-first focus across
it ‘exploit data at scale and speed to make faster, bet- the strategic, operational and tactical levels, and is
ter decisions’ by unlocking data from legacy opera- autonomous by default implementation. These princi-
tional systems, fusing operational and sensor data for ples are embodied by the navy’s platform Pyramid. For
decision-making support, enabling mesh networking example, the NavyX Accelerator seeks to transform the
among different operational systems and sharing data adoption and integration of cutting-edge technologies
across the enterprise.166 Similarly, in 2021, the Ministry in defence through an extensive testing and experi-
of Defence awarded a three-year contract to Anduril to mentation ecosystem. Since scrapping the NELSON
support the implementation of the TALOS programme, programme in 2021, the navy has transitioned to
which seeks to trial an AI-enabled advanced base pro- building a Software House under the ‘Data and Navy
tection system.167 Applications’ initiative. Launched in 2022, the initiative
The army is implementing several simultaneous test- seeks to deliver new reusable capabilities based on a
ing and experimentation projects, including the Human common standard; build a common data fabric for the
Machine Teaming project, which focuses on the integra- navy; and enable an end-to-end electronic workflow
tion of robotics and autonomous systems (RAS), Project between existing legacy capabilities to ensure more util-
WAVELL, and Project Mercury. Together, the projects seek ity and functionality is derived from existing invento-
to incorporate agile software and capability-development ries. Supported by the navy’s Defence Transformation
pathways to accelerate innovation uptake in the armed Fund, the navy is experimenting with intelligent-ship
forces, deliver a brigade-level component enhanced with designs that build on electronic workflows between
robotic systems and an integrated sensor-decider-effector sensors, human deciders and effectors, autonomous
chain by 2025, and evolve new tactics and operational con- mine countermeasure systems, and littoral strike ships.
cepts to deliver maximum deterrence and fighting power. NavyX is deploying a new testbed ship, the XV Patrick
These projects use shorter iterative cycles of one to two Blackett, for testing and experimentation of new digital
years to experiment with and test new open-architecture- and software ship environments.168 Royal Navy sources
based capabilities and strict government-owned data. The are optimistic the ship may soon become its first digital-
Human Machine Teaming project incorporates software-as- experimentation facility for autonomous capabilities.169
a-service and data-management-as-a-service approaches The Royal Marines’ ‘Autonomous Advance Force’ initia-
to deliver human–machine teams on the battlefield, and tive has repeatedly experimented with a variety of sur-
Project Mercury is focused on developing next-generation face, air and land tactical uninhabited vehicles.170
autonomous land capabilities across the support and com- In addition, in 2022, Defence Digital deployed the
bat spectrum. The army’s ‘Land Intelligence, Surveillance, Defence DevSecOps Service (D2S) ‘to accelerate the
Target Acquisition and Reconnaissance’ programme seeks delivery of common platforms and services to enable

31 The International Institute for Strategic Studies

data, digital and AI exploration’, and the Ministry of critical trade-offs are made between introducing new
Defence updated its R&D framework to restructure capability programmes and sustaining existing, albeit
grants around shorter project development and moni- possibly problematic, ones.178
toring phases.171, 172
After the COVID-19 pandemic, The Ministry of Defence is already using software-as-
UK innovation agencies like the Defence Science and a-service contracts for advanced analytics or for service-
Technology Laboratory (Dstl) and Defence and Security wide data management. However, this approach could
Accelerator (DASA) increased the frequency of chal- be further scaled across the defence establishment. This
lenges, competitions and innovation awards, making is all the more important since the Ministry of Defence
them bi-monthly. will continue to struggle with severe shortages of human
Furthermore, UK testing and experimentation in capa- skills for rapid software and AI/ML development in the
bility development and operational-concept development medium term. The ministry is working on an AI skills
across UK Strategic Command and the services greatly strategy, though its effects could take as long as a dec-
benefit from enhanced cooperation with the US and also ade to materialise.
with other allies in NATO and the Indo-Pacific. For exam- Further work is needed on the implementation of
ple, the army’s software-defined innovation efforts are sig- the Data Strategy for Defence to maximise utility out
nificantly enhanced by its participation in the US Army’s of defence data. Currently, defence staff are struggling
Project Convergence, the Royal Navy’s digital-transfor- to cope with severe data-storage constraints and a lack
mation agenda is supported by cooperation in the ‘Tech of clear understanding of how to assess data to ensure
Bridge’ initiative with the US Navy and collaboration with valuable data is not automatically discarded before it is
its Task 59 in the Gulf, and the Royal Air Force has benefit- fully exploited.179
ted from training on the US Air Force’s advanced surveil- Furthermore, greater transparency is needed around
lance system, the GHOST Mk4 UAS.173, 174, 175 risk assessments of software and AI components in
defence capability-development projects. Clearer deliv-
4.1.5. Further challenges ery timelines for software-defined capabilities, enabled
While these are positive steps in the right direction, by the application of agile development procedures, are
more time is needed to assess whether this transition also needed. Closer alignment is also necessary between
is a significant step away from waterfall capability- initial efforts towards agile software and capability
development processes towards more agile, itera- development and modernised approaches to testing,
tive and user-driven ones. For each of the positive verification, validation and certification methodolo-
examples discussed above, there are just as many gies to avoid significant challenges in field testing and
complex UK capability-development programmes experimentation across the services.
that continue to treat software as an adjunct to mili- Finally, the Ministry of Defence is trying to engage
tary hardware because the UK continues to pursue more with start-ups and small- and medium-sized
a capability-based approach. Ajax is the latest and enterprises that develop cutting-edge advanced technol-
most pertinent example.176 The Ministry of Defence ogies. However, there are important concerns around
has similar problems with the Watchkeeper UAS, not the sustainability of resilient ecosystems for dual-use
least because of software design and incomplete and defence-specific technologies. The UK national eco-
requirements and specifications.177 system is significantly smaller than the United States’
Furthermore, agile and iterative software- and and China’s, and even though the UK remains the most
capability-development practices are integrated attractive tech ecosystem in Europe, it is far from com-
mainly in newly launched or unfunded capability plete and resilient in all its sectors. Space is a very good
programmes, and compressed delivery times for new case in point, but quantum, directed energy and bio-
capability and functionality are yet to be made clear. technologies are at risk of following in its footsteps. This
This creates added risks to projects and agile develop- has implications for the ministry’s continuous ability
ment practices, as budgetary constraints demand that to draw on its technological-industrial base in critical

Software-defined Defence: Algorithms at War 32

sovereign-technology areas. It may equally require a ‘continued investment in technological capabilities to
reconsideration of current defence R&D practices to exploit the ever-expanding volume of data, in order to
make innovation competitions easier and more attrac- share relevant information with decision-makers and
tive for a wide range of private-sector actors, as well as action-takers in a timely manner’.184 Furthermore, the
to incentivise closer cooperation between traditional document highlights that ‘technological equalisation
and non-traditional private-sector partners in deliver- helps to make numbers important again’, and that by
ing advanced capabilities to the warfighter. sometimes taking a more agile approach and by sheer
weight of numbers, France’s ‘strategic competitors
4.2. France have the capacity to tip the regional balance, such as
France was the first European nation to publish a Iran in ballistics’.185
defence AI strategy in September 2019 and to prioritise This alignment in relation to emerging technologies
the importance of data and new technologies, notably extends to the importance of moving towards strate-
AI and quantum, for the future of warfare. The link gic autonomy – and defence-technological autonomy
between data and AI was clearly stated in the strategy. – in a European context as well as enhancing the role
The document established a new framework for govern- and contribution of the French armed forces to NATO.
ance of data within defence and identified the following Advanced technology solutions, agile procurement
priority areas for AI use cases: decision support in plan- and better integration of innovative technologies that
ning and execution, collaborative combat, cyber defence ensure interoperability with allies and partners and are
and influence, logistics, support and operational readi- developed in a whole-of-government approach are key
ness, intelligence, robotics and autonomy, and admin- to restoring European and French superiority in the
istration and health. While specific technologies are not warfighting domain.186
an end in themselves, the French armed forces seek to To this effect, in May 2020 the Ministry of the Armed
use AI for ‘ensuring that the armed forces’ decision- Forces issued an official guideline related to defence
making processes have the necessary operational supe- innovation, focusing on attracting and integrating new
riority to give them the upper hand over many types technologies into defence applications. The document
of adversary’.180 Importantly, the document identifies a organisationally restructured R&D processes within
pressing need to achieve technological sovereignty in defence and laid new foundations for R&D governance,
a global technological landscape of new technologies, actions and funding. The annual Document de Référence
such as AI, that is dominated by the US and China. de l’Orientation de l’Innovation de Défense, which features
data and AI-driven capabilities among its highest pri-
4.2.1. Policy alignment orities, set goals for both long-term, planned innovation
There is a strong alignment between French policymak- that is needed to prevent technological surprise, and
ers and military leaders regarding the role of emerging for the short-term, open innovation intended to capture
technologies in ensuring Paris retains its autonomy technological advancements in the private sector and
over assessment and decision-making in defence mat- adopt them in defence applications.187, 188
ters. The ability to exploit data and advanced digi-
tal technologies (e.g., nuclear and space) is widely 4.2.2. Organisational adaptation
acknowledged as a prerequisite for French geopoliti- The French focus on R&D and dual-use technologies orig-
cal relevance, military-technological superiority and inating in the private sector has been accompanied by a
defence-industrial competitiveness.181, 182, 183 series of organisational transformations within the Ministry
The 2022 ‘National Strategic Review’ emphasises that of the Armed Forces intended to reinforce the R&D sec-
in ‘an increasingly competitive and complex interna- tor and provide the French armed forces with access to
tional context, France must focus its efforts on raising its a sovereign supply of critical technologies for defence.
level of knowledge, its appreciation of competitors and These have included the establishment of the Defence
adversaries, and anticipating their intentions’ through Innovation Steering Committee (Comité de Pilotage de

33 The International Institute for Strategic Studies

l’Innovation de Défense) and the Innovation Acceleration In April 2021, the Ministry of the Armed Forces
Standing Committee (Comité Permanent d’Accélération launched the Digital Defence Agency (l’Agence du
de l’Innovation) to manage and steer defence-innovation Numérique de Défense, AND). The new agency is
priorities and activities across the ministry, as well as the responsible for managing ‘complex or high-stakes
Defence Artificial Intelligence Coordination Cell (Cellule digital projects’ within the framework of the ministe-
de Coordination de l’Intelligence Artificielle de Défense) rial policy relating to digital technology defined by
to coordinate all defence projects related to the develop- the Directorate-General for Digital Affairs (Direction
ment, adoption and integration of AI in defence. Générale du Numérique, DGNUM).192
Furthermore, in September 2018, President Macron The newly created structures are focused on ena-
established the AID under the Directorate General of bling the French transition to a system-of-systems
Armaments (DGA), the ministry’s capability-development approach to defence digitalisation and innovation.
and procurement powerhouse. The AID was tasked with The digitalisation of defence in the French context
coordinating innovation across the defence enterprise and has incorporated elements related to the digitalisa-
with the civilian sector to better attract and exploit dual- tion of military equipment, systems and networks,
use and general-purpose technologies for defence applica- as well as the progressive incorporation of new and
tions. The AID implements four types of projects focused advanced technologies like AI in defence applications.
on driving innovation: Furthermore, the ministry has established an ethical
 defence technologies projects: those that refine the committee to oversee the responsible development
technologies necessary for military requirements and use of key advanced technologies in defence,
and are the main vector for planned innovation; including AI, quantum, biotechnologies and human
 innovation-acceleration projects: those that cap- enhancement. The overarching logic of this process is
ture innovations from the civilian sector, adopt in line with foundational tenets of software-defined
them for military use and further develop them defence, notably data, software and AI/ML algorithms.
with ministerial support. Innovation-acceleration
projects represent the second-largest number of 4.2.3. French focus on high-intensity warfare
projects under implementation by the AID (134 and software-defined defence
out of 454 projects);189 The French approach to defence innovation has evolved
 participative innovation projects: those originat- during the last five years. Importantly, the French
ing from French military personnel and focused armed forces’ approach to future conflict has shifted
on specific use-cases at earlier stages; in 2022 towards a prioritisation of high-intensity war-
 research projects: those geared towards long-term fare and away from its previous focus on low-intensity
research and future strategic technologies, usually and counter-terrorism warfare. The transition implies a
in partnership with academia, research organisa- move away from innovation and modernisation of the
tions, schools or companies. As of 2021, research ‘middle segment’ – in which the French military is pre-
projects represented the largest share of the projects pared to address conflicts and threats across the low-
implemented by the AID (169 out of 454 projects).190 to high-intensity, state and non-state threat spectrum
However, the AID has faced strong criticism for the by maintaining a full-spectrum military capability –
mismanagement of these projects and for failing to towards high-intensity warfare.193, 194 This is a require-
rapidly contract projects developing new technolo- ment of the current period of renewed and intensifying
gies for defence applications.191 Though contracting great-power competition, in which technological and
should take 90 days at most, the AID’s timelines information superiority within multi-domain opera-
extended to well over nine months in 2020 and 2021; tions are the markers of battlefield competitiveness.
 in addition, the AID coordinates La Red Team, which High-level military officials have publicly acknowl-
is tasked with developing defence-relevant foresight edged that France’s focus on the middle segment in the
scenarios of future conflict and strategic competition. 2000s and 2010s led to a depletion of specific capabilities

Software-defined Defence: Algorithms at War 34

even as the force remained highly expeditionary. This in the case of the UK, is often risk-averse and depriori-
focus also ossified the long-term planned innovation tises disruptive innovation.200 However, it is a step in
pillar of French R&D to such a degree that in 2021 over the right direction for the armed forces to align threat
95% of innovation projects related to emerging technol- assessments and capability priorities, and to seek to
ogies like AI and data science were linked to existing accelerate the adoption of new technologies in defence,
programmes of record for next-generation capabilities, not just as part of the next-generation capabilities to be
such as the FCAS.195 France’s focus on planned and deployed in the mid- and late-2030s but also within the
incremental innovation, in which new technologies broader armed forces.
are adopted as part of next-generation capabilities or
mid-life upgrades of legacy platforms, created an envi- 4.2.4. Software-defined defence and
ronment in which emerging technologies best succeed French capability development
when tied to existing programmes.196 The Ministry of the Armed Forces already deploys data sci-
The result was an innovation cycle in which tech- ence and AI applications for advanced analytics, predictive
nologies that were not linked to existing programmes maintenance and other tasks. For example, France is using
risked being underfunded or deprioritised and software-as-a-service solutions for the exploitation and anal-
deemed to have a low technology-readiness level, ysis of data generated by its space-based assets, AI-enabled
which could adversely affect the continued R&D of predictive maintenance and munitions health solutions,
such technologies despite their potential use in future sensor-data fusion and analysis for the Rafale fleet, as well as
warfighting operations or for future modernisation predictive big-data analytics and digital-twin solutions for
needs. Paradoxically, technologies linked to high- the Falcon.201, 202, 203 In addition, the French armed forces are
profile capability projects, like FCAS and SCORPION, already deploying and experimenting with a wide range of
were often prioritised despite lower maturity lev- uninhabited land and aerial vehicles.204
els, leading to deadline and cost overruns. In theory, Paris has also implemented several flagship projects
the ministry restructured administrative processes that leverage data, software and AI/ML in defence capa-
within platform-driven capability-development pro- bilities. One of them is ARTEMIS managed by the AND,
grammes, enhanced programme monitoring to ensure which is building an integrated and sovereign digital
incremental integration of technologies in capabili- backbone for the French armed forces, equipping them
ties, and sought greater collaboration with end users with the ability to flexibly fuse and exploit big data, and
to facilitate a genuine system-of-systems approach.197 use a cloud-native architecture and software factories.
However, most efforts towards defence innovation The project was launched in 2017, with the first year-long
and digitalisation in AID, for example, placed more phase seeing three concepts proposed by Capgemini,
emphasis on supporting the civilian-industrial base Atos, Thales and Sopra Steria competing against each
for AI development, among other technologies, than other. In this phase the requirements included a core
on facilitating their rapid adoption and integration execution environment; adequate computing and stor-
into French military systems.198 age capacity; a software factory and a sandbox for the
A reorientation towards high-intensity inter-state development and testing of new software and algo-
conflict carries important implications for the approach rithms on real data in a controlled environment; and a
to capability development and procurement as well software-development kit. And the ministry appeared
as for France’s overall preparedness, as French mili- to adopt an agile development model that emphasised
tary officials acknowledged in 2022 (prompted among early collaboration with the end users and a modular
other reasons by the ongoing war in Ukraine).199 This approach to building and upgrading the capability.
transition is still too recent to allow an assessment of In phase two, launched in 2018, Atos and Thales
its impact on the French embrace of software-defined were chosen to mature and demonstrate their con-
defence or to determine whether it has changed the cepts in six different use-cases: processing of het-
ministry’s organisational innovation culture, which, as erogenous big data; cyber security; staff-health

35 The International Institute for Strategic Studies

monitoring; technical and economic intelligence; 4.2.5. Further challenges
maintenance; and fleet monitoring.205 An interim The French Ministry of the Armed Forces has priori-
capability version of ARTEMIS’ technology demon- tised elements of a software-defined defence approach
strator is already in use with the French armed forces in their vision of exploiting data, AI/ML and digital
and use-cases are subject to several pilots testing the solutions in defence. However, several concerns per-
capability. During phase two of the project, market- sist. Notably, though budgetary allocations for the
consolidation dynamics led Atos and Thales to estab- development and implementation of digital infrastruc-
lish a joint-venture company, ATHEA, which became ture for modernised legacy systems or for new capa-
the sole designated contractor for ARTEMIS in phases bilities have increased, reaching EUR2bn (USD2.4bn)
two and three, the latter launching in mid-2022.206 in 2023, many key digital-infrastructure projects are
Capgemini and Sopra Steria became subcontractors delayed in comparison to the capability-development
to the project. Phase three foresees the industrial roll- timelines. Examples include digital infrastructure for
out of the technology across the ministry and the mil- the ARTEMIS, SCORPION and Rafale upgrade pro-
itary services by 2028. grammes.210 This could lead to delays in the opera-
As the armed forces transitioned towards prioritis- tional deployment of new functions and capabilities.
ing high-intensity warfare in 2022, other ongoing pro- Moreover, obvious gaps in French digital-infrastructure
jects that focus on the AI-enabled exploitation of big planning – such as the lack of any ongoing projects to
data and autonomous capabilities have gained renewed deploy 5G technologies in defence applications – could
importance in the French defence ecosystem.207 Notable pose a different type of challenge to the adoption of AI
examples include SCORPION, which is recapitalising and more advanced software in defence, even as France
the lighter segment of land capabilities in the army with is the European leader in satellite-based communica-
a modernised C2 and communications system; TITAN, tions and intelligence.
which is modernising the army’s high-intensity-warfare Official sources have presented the flagship capability-
capabilities for high-connectivity, networked, multi- development projects discussed above as examples of
domain combat; and VULCAIN, which seeks to expand agile integration of new technologies into defence capa-
the number of autonomous land logistics and combat bilities, based on modular, open architectures. However,
vehicles in the army every few years. In the air domain, analysis of these programmes suggests they are devel-
the FCAS and ‘Man-Machine-Teaming’ programmes were oped through a classic waterfall capability-development
developed in coordination to provide a sixth-generation model, where software development is often separated
manned and uninhabited air-combat capability for multi- into different strands. While this is the case for the
domain operations. The programmes are already develop- SCORPION and FCAS projects, ARTEMIS elicits similar
ing several AI-enabled capabilities for data fusion as well concerns. Important questions remain unanswered about
as analytics for enhanced situational awareness and moni- the full capabilities and functionalities of ARTEMIS in
toring of pilot health. Furthermore, in April 2021 the min- phase three, solutions for substantial data and communi-
istry launched the BRASIDAS project, which is expected to cations challenges across the enterprise, the overall cost of
deliver an AI-enabled predictive maintenance solution for the contract, and the general approach towards upgrades,
the H225M Caracal helicopter and Patroller drone fleets.208 maintenance and new functionality integration, which are
The project is also expected to extend to vertical-lift fleets, key to a software-defined defence approach. Therefore, it
such as France’s AS532UL Cougar and Tiger helicopters is unclear whether the Ministry of the Armed Forces is
and the Rafale and Mirage 2000.209 Project SIBIL (Système genuinely moving towards more agile, iterative and mod-
d’information pour la prévision des besoins et l’innovation logis- ular approaches to defence-capability development.
tique), which uses algorithms for the predictive mainte- Furthermore, it seems the French defence establish-
nance of land vehicles, is under experimentation with the ment is being pulled in different directions by different
French army. All these projects will be enabled by a com- institutional interests. France’s 2019 defence AI strategy
bat cloud foundation. sought to increase the defence establishment’s appetite

Software-defined Defence: Algorithms at War 36

to develop capabilities based on new technologies by country’s technological and data sovereignty. For exam-
establishing a formal process to monitor whether they ple, Germany’s 2020 Artificial Intelligence Strategy
are trustworthy and responsibly used by the armed emphasises key societal use-cases in the economy,
forces. However, the organisational culture within the health and industry, the need for action that consoli-
French defence establishment remains risk-averse and dates AI skills in society and consolidates national and
oriented towards long-term incremental innovation and cross-border European AI ecosystems, and the foster-
technological uptake. While this is also true of other ing of an ethical approach towards the development of
European countries, in France there is a significant dis- trustworthy and responsible AI applications.213
connect between the urgency of the policy narrative Furthermore, the country’s 2021 Data Strategy under-
around the technological and digital modernisation of lined the need to improve data provision and secure data
the armed forces and, in practice, the misalignment of access at the infrastructural level, to promote responsi-
innovation and procurement. ble data use and tap potential for innovation, to improve
data skills and establish a new data culture in Germany,
4.3. Germany and to make the Federal Government a world leader
In 2018, Germany’s then-minister of defence Ursula von of the new data culture. The document highlighted the
der Leyen acknowledged that ‘every “battle”, whether potential for gaining geo-economic competitiveness
it is fought on land, at sea or in the air, is at the same by incentivising data-centric innovation, including in
time always a battle for “information power”, which is new technologies such as quantum computing, while
why armed forces such as the Bundeswehr need their expanding Germany’s technological sovereignty and
own networks and software to be both functional and establishing a national regime for data security and
resilient’.211 She also acknowledged the need to ‘develop protection by design. The Data Strategy announced the
processes in order to securely and profitably use the government’s intention to create a shared data pool that
exponentially growing amount of information and data allows for a standardised, accessible data format for use
we can collect today thanks to modern technology’.212 by any federal authority, although standardisation was
This acknowledgement is not unique among the German acknowledged to be a substantial challenge.214
political and military elite, where there is wide accept- By contrast, there is no political consensus around
ance of the need for the Federal Government and the concrete actions towards the digitalisation of defence
armed forces to digitalise in order to take advantage of and the adoption and integration of emergent tech-
and prepare for emerging and disruptive technologies. nologies like AI/ML, despite a high-level policy pri-
While the need for greater digitalisation of the German oritisation of a data-driven approach. The Ministry of
armed forces has become greater over the last five years, Defence adopted a Defence Data Strategy in 2021, in
concrete actions have not been forthcoming. As a result, line with multinational developments in NATO and
out of the five case-study countries, Germany remains the EU.215 The document set the objectives of establish-
the furthest away from embracing a software-defined ing a defence data-governance framework and a stand-
defence approach. ardised data infrastructure; achieving a data-oriented
organisational culture and increased institutional data
4.3.1. Policy alignment towards a society- literacy; enhancing innovative use of data, ensur-
first approach ing data quality and availability across defence; and
The lack of concrete action towards the digitalisation of enhancing the operational resilience and readiness of
defence and adoption of advanced technologies like AI/ IT networks and weapons systems while reducing their
ML in defence stands in harsh contrast to the Federal life-long costs and enabling big-data analytics.216 Other
Government’s policy interest and investment in emerg- service-level policy guidance on defence data manage-
ing technologies and data exploitation in a civilian, ment, like the Bundeswehr’s digitalisation strategy,
economic and industrial context. Indeed, there is broad remains classified. The ministry is reportedly imple-
political consensus in Germany around increasing the menting over 1,000 digitalisation projects, based on

37 The International Institute for Strategic Studies

a phased ‘connecting-islands-of-digitalisation’ plan, of Defence are built on reusable standard building blocks
many of which are component- and service-level initi- to enhance interoperability and diminish life-long costs.
atives facing long delays due to legal and bureaucratic However, the focus of the ministry’s digitalisation activi-
bottlenecks.217 Explaining the Bundeswehr’s incre- ties remains the digitalisation of the defence enterprise
mental approach to digitalisation, the digital officer and business systems, with measures to increase the
for land-based operations said this: ‘We have to build digitalisation capacity of weapons systems ‘still in their
small, manageable islands, fully digitalise them, bring infancy’ according to German official sources.221 In 2021
about success, and then carry that success over into the ministry was managing 81 digitalisation activities,
other areas by creating new islands. We are not start- with implementation timelines mostly of one to four
ing with full divisions or brigades, so not with 30,000 years, or longer in a few cases.222
or 10,000 people, but rather with a battle group. That
is around 1,500 people and 800 vehicles. And those are 4.3.2. German software-defined
going to be fully digitalised in the system.’218 There is defence innovation
no public evidence of the ministry moving towards The German armed forces operate several platforms and
enterprise-wide system-of-system approaches con- types of ammunition with some or full autonomous capa-
sistent with multi-domain operations and a software- bility. This includes the autonomous functions of aid-
defined defence approach. defence systems, UAVs such as the Puma 3 AE tactical
Moreover, the 2020 updated Strategy Paper of the UAV, and the LUNA UAV, the Harop loitering munition,
Federal Government on Strengthening the Security uninhabited ground vehicles (UGVs) like the THeMIS tac-
and Defence Industry featured four digital technolo- tical UGV, and other autonomous counter-mine clearance
gies out of the total eight key technologies identified as and underwater autonomous ISR capabilities.223 In addi-
important for the future of the German armed forces: tion, Germany participates in the FCAS project which, as
security-relevant IT and communication technologies, discussed in the previous subsection, incorporates sev-
AI, sensors and network-enabled operations/crypto eral software-defined elements. However, Germany has
technologies.219 A research branch of the Bundeswehr opted out of an uninhabited version of FCAS.
emphasised the need to adopt networked military capa- In addition, the Ministry of Defence is already
bilities and achieve mass and manoeuvrability as fun- using an AI-enabled early-warning system from crisis
damental measures to prepare for the future of warfare management as well as AI-enabled military person-
and an increasingly transparent battlespace covered by nel-health and travel applications, with other projects
smart networks of active and passive sensors.220 currently under development.224, 225
Meanwhile, the German armed forces are struggling Digitalisation and AI adoption is provisionally part
with severe connectivity and communications challenges, of other ongoing modernisation efforts. One example is
where legacy systems cannot share data either with each the HERKULES follow-up project, which enabled a basic
other or with close allies. The prioritisation of the digi- cloud infrastructure for the Bundeswehr and the digi-
talisation of land forces is linked to the TEN and D-LBO talisation of its healthcare databases and services to per-
projects, which are interdependent and driven by multi- sonnel.226 Another is the MESBw, which is modernising
national requirements for NATO and EU missions and command, control, communications, computers and intel-
operations. For example, project TEN is modernising and ligence (C4I) software and infrastructure for the armed
digitalising tactical communications and data-exchange forces based on an open architecture model that intercon-
capabilities between the German and Dutch land forces, nects legacy proprietary systems, such as command and
whose military equipment was not previously interop- weapon-deployment systems (FüWES) and weapon-sys-
erable. In 2019 the Ministry of Defence launched SysZ tem platforms.227 Since 2018, the special-operations forces
Digla, the digital organisational and functional element have successfully tested the ‘Multi-sensor real-time com-
of the digitalisation of land forces, through the D-LBO bat in an ad hoc mesh network’ (MEGA) prototype, while
and TEN projects. All digital projects within the Ministry other units in the Bundeswehr have tested Distributed

Software-defined Defence: Algorithms at War 38

Cyber Reconnaissance as a Service (DCR) prototypes.228 not have sufficient organisational and bureaucratic pull,
The ministry is also slowly rolling out the Harmonisation financial strength or embeddedness with end users to
of Management Information Systems (HaFIS) project to drive ambitious innovation agendas and projects, even
modernise the mission-planning tools of stationary and if their capacity and knowledge base is reasonably solid.
deployable command posts. As the ministry is seeking Overcoming historical legacies and a risk-averse
commercially available, off-the-shelf software within the and overly bureaucratic organisational culture in the
aforementioned projects, the largest risks associated with Ministry of Defence will remain a severe challenge,
the projects stem from the lack of supporting infrastruc- as shown by recent issues around the procurement
ture and digital skills, rather than from low levels of tech- of equipment as part of the EUR100bn (USD109bn)
nological maturity as in the cases of France, the UK and fund. At stake is not just the absorption of the one-time
the US.229 EUR100bn (USD109bn) fund, which depends on deep
Most ministry and Bundeswehr innovation initiatives or reforms in Germany’s defence-procurement framework,
strategies for data-driven, software-defined or AI-enabled but also ensuring a balanced approach between off-the-
capabilities are tied in with Germany’s broader goal of shelf purchases that deliver new equipment quickly,
securing EU defence initiatives and digital sovereignty, or an understandable impulse to spend on domestic and
take a back seat to existing NATO goals, development strat- European industrial capacity, and increasing invest-
egies or modernisation/innovation requirements. Existing ment in defence R&D, including greater investment in
efforts include software-defined radios as part of the EU emerging technologies.234 Longer-term German pros-
European Secure Software defined Radio (ESSOR) project and pects for embracing a genuine software-defined defence
some development of AI-enabled military systems, such as approach depend on a shift away from current ossi-
the FCAS programme with France and Spain. fied procurement practices to deliver new functionality
Little innovation is practised exclusively by Germany faster to the warfighter.
for its own benefit.230 The ministry’s most recent Military Moreover, German defence experts have expressed
Scientific Research Report shows that fewer than 10% of concerns that the country’s defence establishment is too
the projects focus on digitalisation (15 out of 158) and focused on digitalisation (digital and cyber innovation)
fewer than 4% on AI. Instead, all AI projects are exclu- at the expense of other critical components of future
sively defensive in nature and focused on health, ISR and warfare, notably robotics and automation, and a more
other tasks rather than enhanced warfighting capacity.231 flexible approach to software-defined capabilities.235
This is rather paradoxical given the degree of scientific Germany’s coalition government has announced an
expertise in the fields of AI and other emerging technolo- intention ‘to initiate a streamlining of public procure-
gies that exists in Germany. However, it is indicative of ment law’ to simplify and accelerate defence procure-
the highly politicised nature of defence investment and ment or exempt key technologies, including advanced
defence modernisation in the country.232 digital technologies, from complying with Germany’s
strict public-procurement law.236
4.3.3. Further challenges However, there is currently no indication that Germany
Germany will continue to be domestically challenged in is moving away from very lengthy capability-development
its ambition to fully embrace software-defined defence and procurement programmes or that it is pursuing more
in the short and medium term. Despite solid German agile and iterative software development and integration
support for NATO and EU solidarity and interoperabil- projects, as recently exemplified by its participation in
ity in security and defence, and despite the ongoing war the F-35, THeMIS tactical UGV and FCAS programmes.
in Ukraine, Berlin still lacks both a clear threat assess- This is partially explained by the fact that Germany, to a
ment to inform capability development and a convic- greater degree than France, is still attempting to fill his-
tion of its responsibility to defend.233 torical capability gaps, rather than thinking dynamically
New German defence-innovation structures, such as about modern warfare and how its armed forces can cred-
the Cyber Innovation Hub and the Cyber Agency, do ibly deter, fight and win.

39 The International Institute for Strategic Studies

5. Conclusion

The research that underpins this report focused on hardware-embedded, which means it cannot be easily
three core tasks. Firstly, it aimed to conceptualise upgraded without simultaneous hardware upgrades.
software-defined defence as an emerging yet funda- This limits defence establishments’ data rights in using
mental architectural, organisational and operational the software, retrofitting legacy systems with new soft-
principle of modern military operations. It explored ware solutions, and upgrading the capability frequently
the transformational role of data, software and AI/ML for increased functionality. Some examples of core com-
in defence applications today and in the near future. mon operating systems are beginning to emerge each of
In doing so, the paper explored four underpinning the three countries, but they remain the exception rather
elements of a software-defined defence approach: than the rule. There is no TankOS or FighterOS on the
a changing relationship between military software horizon at the moment, as defence software remains
and hardware that means technological progress highly fragmented and often lacking interoperability
is faster in software and the promise of operational with other service, national or allied systems.
advantage in information superiority is underpinned Lastly, the research explored in detail the national
by software-defined functionality of systems; a data- efforts of the five case-study countries – China, France,
centric approach to developing and architecting new Germany, the UK and the US – towards software-
capabilities and systems-of-systems; a human-centric defined defence. It found that the intensifying strategic
approach to API-enabled end-to-end electronic work- competition between the US and China is accelerating
flows that is designed to enhance human capacity the transition towards software-defined defence in the
and safety; and software as a core consideration for two countries. However, both the US and China con-
weapon and system design and upgrade. tinue to encounter significant challenges in their efforts
Secondly, the paper sought to assess current chal- to achieve superiority over the other in the domain of
lenges in developing and deploying modern, AI-based software-defined capabilities. Yet, despite these chal-
defence software in five case-study countries – China, lenges, Beijing’s sustained efforts towards the digitalisa-
France, Germany, the UK and the US. While there are tion and intelligentisation of defence means the West’s
incipient efforts to integrate agile and iterative software- competitiveness advantage in software-defined defence
defined defence models, particularly in France, the UK is narrowing. The United States’ advantage is increas-
and the US, these are far from generalised. Most defence ingly confined to discrete areas (e.g., financial, technol-
software and AI applications are developed through ogy, net organisational power and adoption patterns).
traditional waterfall models of capability development Nevertheless, the US and China remain ahead of
as an added layer to military hardware, despite these France, Germany and the UK, whose efforts towards
applications in some cases controlling over 80% of the implementing software-defined defence have been
capability’s functionality and performance. Software more modest. The UK and France foster a greater level
is very often considered a moderate or critical risk to of ambition than Germany towards the digitalisation of
capability programmes. Furthermore, the structure of their armed forces and the incorporation of advanced
defence-procurement contracts and property rights technologies like AI for operational and information
continues to pose significant challenges when migrat- advantage. London and Paris have developed the strate-
ing defence software to modular and open architec- gies, organisational structures and financial tools to pur-
tures, and for the development and integration of AI sue a software-defined defence approach. However, the
applications. Most deployed or in-development defence use of agile and iterative software development remains
software in France, the UK and US is customised or too timid in both cases. France and the UK are therefore

Software-defined Defence: Algorithms at War 40

still in the early stages of a transition to software-defined There are clearly similar challenges to transitioning to
defence, lacking a firm commitment to this transforma- software-defined defence both in the US and in Europe.
tion for a combination of bureaucratic, organisational- However, they remain a higher barrier to entry for the
culture, financial and industrial reasons. Europeans. Digitalisation of defence is more expensive
By contrast, there are no indications that Germany in Europe than in the US because of the high level of
is embracing a software-defined defence approach. fragmentation of military equipment and entrenched
Instead, Berlin’s capability-development plans and defence-industrial interests around proprietary hard-
procurement frameworks remain ossified around ware and software. Beyond the siloed data, European
long-term waterfall development models and off-the- defence ecosystems still lack fundamental building
shelf procurement of proprietary systems incorporat- blocks for software-defined defence, including a solid
ing bespoke software linked to proprietary hardware. governance of AI-enabled autonomy, robust enabling
Efforts towards achieving digitalisation of defence and infrastructure and better leverage of alternative sources
adopting a data-centric approach are slow and easily of funding (e.g., capital markets).
bogged down in bureaucratic procedures. This report finds that a transatlantic gap in software-
While software-defined defence entails a focus on hori- defined defence (capability and doctrinal/operational)
zontal hyperscaling of infrastructure, functionality and per- has already emerged. Compared to the Europeans,
formance, defence establishments that are natural vertical the US is more advanced in the technological, fund-
hyperscalers in conventional military capabilities continue ing, planning, experimental and doctrinal aspects of
to struggle as microscalers of software-enabled defence. software-defined defence. US software-defined defence
As the European continent is shaken by high-intensity is and will remain much more scalable and better
conventional war in Ukraine, early lessons from the bat- funded than European efforts. However, the mounting
tlefield suggest the importance of flexible approaches to challenges in the United States’ adoption of AI/ML in
the integration of innovative software solutions. China, defence mean the transatlantic gap is relatively narrow,
the UK, the US and France are closely watching the despite the scale and speed of US efforts. Therefore, it is
developments on the battlefield and drawing insights our assessment that the transatlantic software-defined
for the future of high-intensity inter-state war. The war capabilities gap is still bridgeable in the medium to long
has spurred another wave of interest in defence invest- term if the Europeans accelerate their efforts and mus-
ment across Europe, in NATO and the EU as well as in all ter the political will to fund the development of modern
three of the European nations analysed here. Berlin, how- defence capabilities. The lessons learned from the ongo-
ever, continues to face substantial challenges in adapting, ing war in Ukraine could be an important catalyst for
more so than any of the other nations discussed here. this transformation.

41 The International Institute for Strategic Studies

Notes

1 Christian Brose, The Kill Chain: Defending America in the Future Cost%20Estimation%20Metrics%20Manual%20for%20
of High-Tech Warfare (New York, NY: Hachette, 2020), p. 61. Defense%20Systems.pdf.
2 Mark E. Nissen, ‘JSOW Alpha Contracting Case Study 8 Christian Hagen et al., ‘Software: The Brains Behind U.S.

(Software Version)’, US Naval Postgraduate School, 1997, Defense Systems’, ATKearney, 2012, p. 2, http://ael.chungbuk.

https://man.fas.org/dod-101/sys/smart/docs/jsowcase.htm. ac.kr/ael/ref/military_technologies/software/hagen(xx)-

3 Andrew Philip Hunter, Schuyler Moore and Maura Rose Software-The_Brains_Behind_US_Defense_Systems.pdf.

McQuade, ‘Acquisition of Software-defined Hardware- 9 Richard Hanna et al., ‘Innovation Timelines from Invention to

based Adaptable Systems’, Center for Strategic and Maturity: A Rapid Review of the Evidence on the Time Taken for

International Studies, 7 August 2019, p. 4, https://www.csis. New Technologies to Reach Widespread Commercialisation’,

org/analysis/acquisition-software-defined-hardware-based- UK Energy Research Centre, December 2015, p. xvi, https://

adaptable-systems. d2e1qxpsswcpgz.cloudfront.net/uploads/2020/03/innovation-

4 See, for example, Nand Mulchandani and Lt. General timescales-working-paper-march-2016.pdf.

(Ret.) John N.T. ‘Jack’ Shanahan, ‘Software-defined 10 Author interview with former military official, July 2022.

Warfare: Architecting the DOD’s Transition to the Digital 11 Megan Eckstein, ‘Navy’s Digital Horizon exercise showcases

Age’, Center for Strategic and International Studies, 6 power of “mesh networks,” AI’, C4ISRNET, 14 December 2022,

September 2022, https://www.csis.org/analysis/software- https://www.c4isrnet.com/naval/2022/12/14/navys-digital-

defined-warfare-architecting-dods-transition-digital-age; horizon-exercise-showcases-power-of-mesh-networks-ai/.

and Jason Weiss and Dan Patt, ‘Software Defines Tactics: 12 Craig Fields et al., ‘Design and Acquisition of Software for
Structuring Military Software Acquisitions for Adaptability Defense Systems’, US Department of Defense, Defense
and Advantage in a Competitive Era’, Hudson Institute, Science Board (DSB), 14 February 2018, pp. 3 and 23, https://
December 2022, https://s3.amazonaws.com/media.hudson. apps.dtic.mil/sti/pdfs/AD1048883.pdf.
org/Software+Defines+Tactics.pdf. 13 Simona R. Soare, ‘Cooperative Edge: Key Drivers of Defence

5 Nissen, ‘JSOW Alpha Contracting Case Study (Software Innovation in Multilateral Organisations Like the EU and

Version)’. NATO’, IISS research paper, forthcoming 2023.

6 National Research Council, ‘Critical Code. Software 14 Emmanuel Huberdeau, ‘Geospatial intelligence: Earthcube

Producibility for Defense’, 2010, p. 19, https://apps.dtic.mil/ promotes automated imagery analysis’, Jane’s International
dtic/tr/fulltext/u2/a534043.pdfhttps://apps.dtic.mil/dtic/tr/ Defence Review, 18 November 2019.

fulltext/u2/a534043.pdf. 15 Ted Johnson and Charles F. Wald, ‘The Military Should Teach

7 Source lines of code (SLOC) are a common measure for AI to Watch Drone Footage’, Wired, 26 November 2017, https://

assessing the size of defence software used by France, www.wired.com/story/the-military-should-teach-ai-to-

Germany, the UK and the US. According to the Naval watch-drone-footage/.

Postgraduate School’s ‘Software Cost Estimation Metrics 16 US Government Accountability Office, ‘Battle Management:

Manual for Defense Systems’ manual, SLOC are ‘logical DOD and Air Force Continue to Define Joint Command and

source statements consisting of data declarations and Control Efforts’, GAO-23-105495, January 2023, p. 18, https://

executables’ which can be classified into new, adapted www.gao.gov/assets/820/814635.pdf.

(reused, modified), equivalent, generated, converted and 17 Ibid, p. 9.

commercial off-the-shelf software. The latter forms the basis 18 German Federal Ministry of Defence, ‘Information from the

for cost assessments for software-development contracts in Parliamentary Commissioner for the Armed Forces: Annual

the defence establishment. See Bradford Clark and Raymond Report 2020 (62nd Report)’, 23 February 2021, p. 93, https://

Madachy (eds.), Software Cost Estimation Metrics Manual www.bundestag.de/resource/blob/839328/e1a864120

for Defense Systems, (Haymarket, VA: Software Metrics Inc., 697c27057534944ceb20111/annual_report_2020_62nd_report-

2015), pp. 19–20, http://www.softwarecost.org/Software%20 data.pdf.

Software-defined Defence: Algorithms at War 42

19 Examples include the digital systems onboard Germany’s 31 Dassault Aviation, ‘Launch of the Man Machine Teaming

Tornado multirole fighter aircraft, key DoD network enterprise Advanced Study Programme’, 16 March 2018, https://www.

systems and C2 and mission-planning tools still running dassault-aviation.com/en/group/press/press-kits/launch-

on Windows 2008 servers, and the DoD’s use of eight-inch man-machine-teaming-advanced-study-programme/; and

floppy disks, which have not been used commercially in over Airforce Technology, ‘France launches Man-Machine-

two decades, for data sharing in its nuclear forces. Teaming Programme to Develop AI for Combat Aviation’,

20 Jed Judson, ‘Project Convergence exercise has new gateway 20 March 2018, https://www.airforce-technology.com/

to test emerging tech’, DefenseNews, 11 October 2022, https:// news/france-launches-man-machine-teaming-programme-

www.defensenews.com/training-sim/2022/10/10/project- develop-ai-combat-aviation/.

convergence-exercise-has-new-gateway-to-test-emerging-tech/. 32 The available open-source information regarding Chinese

21 IISS workshop, ‘Future-proofing Defence: Opportunities capability-development practices in general, and software-

and Challenges of Software-defined Defence’, 7 December development practices in particular, is insufficient to

2022. See also Megan Eckstein, ‘Navy’s Digital Horizon substantiate a clear determination and assessment as in the

exercise showcases power of “mesh networks,” AI’; and case of the four Western case studies investigated here.

Jon Harper, ‘Navy to Establish Additional Unmanned Task 33 The waterfall capability-development model is the most

Forces Inspired by Task Force 59’, Defensescoop, 4 December commonly used framework for major capability programmes

2022, https://defensescoop.com/2022/12/04/navy-to-establish- across the five case-study countries studies analysed in this

additional-unmanned-task-forces-inspired-by-task-force-59/. paper. It has been the predominant framework for capability

22 Jason Weiss and Dan Patt, ‘Software Defines Tactics: development since the 1970s, and relies on strict, linear

Structuring Military Software Acquisitions for Adaptability and interconnected phases in which previous phases must

and Advantage in a Competitive Era’, p. 12. be successfully completed before proceeding to the next

23 Simona R. Soare, ‘Algorithmic Power, NATO and Artificial phase. The phases of a waterfall model include definition

Intelligence’, IISS Military Balance Blog, 19 November of requirements, design, execution, testing and release.

2021, https://www.iiss.org/blogs/military-balance/2021/11/ By contrast, agile models build on small increments of the

algorithmic-power-nato-and-artificial-intelligence. capability requirements and enable a continuous integration/

24 Hayley Patterson, ‘From Software to Battlespace: Reimagining continuous delivery approach. Other models include

How Defence Does DevSecOps with D2S’, Defence Digital/ incremental, spiral, agile, DevOps, and hybrid or mixed. For

Digital Foundry, UK Ministry of Defence, 19 August 2022, a comprehensive definition of these capability-development

https://defencedigital.blog.gov.uk/2022/08/19/from-software- models and how they apply to software, see US Governmental

to-battlespace-reimagining-how-defence-does-devsecops- Accountability Office, ‘DoD Software Acquisition: Status of

with-d2s/. and Challenges Related to Reform Efforts’, GAO-21-105298,

25 National Research Council of the National Academies, September 2021, p. 4, https://www.gao.gov/assets/gao-21-

‘Critical Code. Software Producibility for Defense’, 2010, p. 105298.pdf.

19, https://apps.dtic.mil/dtic/tr/fulltext/u2/a534043.pdf. 34 US Government Accountability Office, ‘Report to

26 Mulchandani and Shanahan, ‘Software-defined Warfare: Congressional Committees: Weapon Systems Annual

Architecting the DOD’s Transition to the Digital Age’, p. 6. Assessment. Challenges to Fielding Capabilities Faster Persist’,

27 Ibid, p. 11. June 2022, https://www.gao.gov/assets/gao-22-105230.pdf.

28 US Navy, ‘Software Ecosystem Architectural Model and 35 These include emerging and emergent technology stacks

Application Program Interface for Common Core Combat in AI/ML, data science, quantum, space, next-generation

System’, Navy SBIR 2020.1 – Topic N201-057, 14 January (tele)communication, biotechnology, human enhancement,

2020, https://www.navysbir.com/n20_1/N201-057.htm. directed energy, new propulsion systems, advanced

29 Mulchandani and Shanahan, ‘Software-defined Warfare: manufacturing, additive materials and more.

Architecting the DOD’s Transition to the Digital Age’, p. 10. 36 See, for example, a comprehensive list of US Department of

30 IISS workshop, ‘Future-proofing Defence: Opportunities and Defence recommendations on adapting software development

Challenges of Software-defined Defence’. practices since the late 1980s in Defence Innovation Board,

43 The International Institute for Strategic Studies

 ‘Software Acquisition and Practices (SWAP) Study’, May 45 Niall McCarthy, ‘The Mammoth Cost of Operating

2019, pp. 19–21, https://innovation.defense.gov/software/. America’s Combat Aircraft’, Forbes, 26 November 2020,

37 See, for example, Leonardo, ‘Leonardo Accelerates the https://www.forbes.com/sites/niallmccarthy/2020/11/26/

Digitalization of the Aerospace Defence and Security the-mammoth-cost-of-operating-americas-combat-aircraft-

Sector. Genoa Is the National Industrial Competence Hub’, infographic/?sh=3e26f1e87da7.

1 December 2021, https://www.leonardo.com/en/press- 46 Simona R. Soare and Fabrice Pothier, ‘Leading Edge: Key

release-detail/-/detail/01-12-2021-leonardo-accelerates-the- Drivers of Defence Innovation and the Future of Operational

digitalization-of-the-aerospace-defence-and-security-sector. Advantage’, IISS, 11 November 2021, p. 23, https://www.iiss.

38 Steven Rosenbush, ‘Big Tech Is Spending Billions on AI org/blogs/research-paper/2021/11/key-drivers-of-defence--

Research. Investors Should Keep an Eye Out’, Wall Street innovation-and-the-future--of-operational-advantage.

Journal, 8 March 2022, https://www.wsj.com/articles/big-tech- 47 Fields et al., ‘Design and Acquisition of Software for Defense

is-spending-billions-on-ai-research-investors-should-keep- Systems’, pp. 1–2.

an-eye-out-11646740800. 48 Ibid.

39 See, for example, Macrotrends, ‘Palantir Technologies 49 Shephard Media, ‘Rheinmetall Joins Helsing to Accelerate

Research and Development Expenses 2019-2022 | PLTR’, 2022, AI for Land Systems’, 15 September 2022, https://www.

https://www.macrotrends.net/stocks/charts/PLTR/palantir- shephardmedia.com/news/digital-battlespace/rheinmetall-

technologies/research-development-expenses. joins-helsing-to-accelerate-ai-for-land-systems/.

40 GlobalData, ‘Top Big Data Patent Holders in the Aerospace 50 Palantir, ‘Palantir Announces Availability of Foundry on

and Defence Sector (2002-2022)’, November 2022, https:// Microsoft Azure’, Palantir Blog, 25 January 2023, https://blog.

www.globaldata.com/data-insights/aerospace-and-defence/ palantir.com/palantir-announces-availability-of-foundry-on-

global-top-big-data-patents-holders-in-the-aerospace-and- microsoft-azure-9120311e2d1a.

defence-sector-2131090/. 51 Atos, ‘AWS and Atos Strengthen Collaboration with New

41 Aerospace, Security and Defence Industries Association of Strategic Partnership to Transform the Infrastructure

Europe (ASD), ‘2022 Facts & Figures’, 2022, pp. 21–23, https:// Outsourcing Industry’, 30 November 2022, https://atos.net/

asd-europe.paddlecms.net/sites/default/files/2022-11/ASD_ en/2022/press-release_2022_11_30/aws-and-atos-strengthen-

Facts%20%26%20Figures%202022.pdf. collaboration-with-new-strategic-partnership-to-transform-

42 Statista, ‘Expenditure on Research and Development the-infrastructure-outsourcing-industry.

of Defense Technology Supplier Lockheed Martin from 52 Office of the Under Secretary of Defense (Comptroller), ‘RDT&E

2002 to 2021’, January 2023, https://www.statista.com/ Programs (R-1)’, April 2022, p. iii, https://comptroller.defense.
statistics/268928/expenditure-on-research-and-development- gov/Portals/45/Documents/defbudget/FY2023/FY2023_

of-defense-supplier-lockheed-martin/. r1.pdf; and US Government Accountability Office, ‘Artificial

43 US Government Accountability Office, ‘F­35 Joint Strike Intelligence: DOD Should Improve Strategies, Inventory

Fighter: Cost Growth and Schedule Delays Continue, Process, and Collaboration Guidance’, GAO-22-105834, March

Statement of Jon Ludwigson, Director, Contracting and 2022, p. 2, https://www.gao.gov/assets/gao-22-105834.pdf.

National Security Acquisitions’, GAO-22-105943, 27 April 53 Douglas Barrie, Nick Childs and Fenella McGerty, ‘US

2022, pp. 10–12, https://www.gao.gov/assets/730/720256. FY2023 defence budget request: something old, something

pdf; and US Government Accountability Office, ‘F­35 Joint new…’, IISS Military Balance Blog, 14 April 2022, https://

Strike Fighter: DOD Needs to Update Modernization www.iiss.org/blogs/military-balance/2022/04/us-fy2023-

Schedule and Improve Data on Software Development’, defence-budget-request-something-old-something-new.

GAO-21-226, March 2021, p. 28, https://www.gao.gov/ 54 Margarita Konaev et al., ‘US Military Investments in

assets/720/713140.pdf. Autonomy and AI: A Budgetary Assessment’, Center for

44 US Government Accountability Office, ‘Tactical Aircraft: Security and Emerging Technology, October 2020, pp. 15–18,

F-22A Modernization Program Faces Cost, Technical, and https://cset.georgetown.edu/wp-content/uploads/CSET-U.S.-

Sustainment Risks’, GAO-12-447, May 2012, p. 6, https:// Military-Investments-in-Autonomy-and-AI-A-Budgetary-

www.gao.gov/assets/gao-12-447.pdf. Assessment-1.pdf; and Chris Cornillie, ‘Artificial Intelligence

Software-defined Defence: Algorithms at War 44

 & Machine Learning: BGOV Market Profile’, Bloomberg January 2023, https://www.gov.uk/government/publications/

Government, 23 September 2020, https://about.bgov. defence-science-and-technology-programmes-and-projects/

com/reports/market-profile-artificial-intelligence-and- ministry-of-defences-science-and-technology-portfolio.

machine-learning/. 65 HM Treasury, ‘Budget 2020: Delivering on Our Promises

55 Authors’ calculations based on Office of the Under Secretary to the British people’, 11 March 2020, p. 85, https://assets.

of Defense (Comptroller), ‘RDT&E Programs (R-1)’. This publishing.service.gov.uk/government/uploads/system/

estimate is based on the enacted FY22 and requested FY23 uploads/attachment_data/file/871799/Budget_2020_Web_

funding for military services and DoD-wide projects in Accessible_Complete.pdf.

which AI/ML either represents a key deliverable, a critical 66 UK National Audit Office, ‘The Equipment Plan 2022 to

component of the deliverable or in which AI/ML is used in 2032’, 29 November 2022, p. 18, https://www.nao.org.uk/

a prominent role in the development process. We recognise wp-content/uploads/2022/11/NAO-Report-The-equipment-

the definitional problem where definitions of AI/ML are still plan-2022-to-2032.pdf.

inconsistent across services and the DoD. 67 Based on UK MoD software services tenders between January–

56 Ibid. December 2022 listed in Bidstats. For more details, see https://

57 Govini, ‘The National Security Scorecard: Critical bidstats.uk/tenders/?ntype=tender#792330239-792068335-47.

Technologies Edition’, 29 June 2022, p. 4, https://govini. 68 UK National Audit Office, ‘The Digital Strategy for Defence:

com/wp-content/uploads/2022/06/Govini-National-Security- A Review of Early Implementation’, 19 October 2022, p. 2,

Scorecard-Critical-Technologies.pdf. https://www.nao.org.uk/wp-content/uploads/2022/10/NAO-

58 Ibid, p. 21. report-The-Digital-Strategy-for-Defence-A-review-of-early-

59 Ashwin Acharya and Zachary Arnold, ‘Chinese Public AI implementation.pdf.

R&D Spending: Provisional Findings’, Center for Security 69 Ibid.

and Emerging Technology Issue Brief, December 2019, p. 13, 70 UK Ministry of Defence, ‘Annual Report and Accounts

https://cset.georgetown.edu/publication/chinese-public- 2021–22’, 14 July 2022, https://assets.publishing.service.gov.

ai-rd-spending-provisional-findings/. uk/government/uploads/system/uploads/attachment_data/

60 Ryan Fedasiuk, Jennifer Melot and Ben Murphy, ‘Harnessed file/1090612/20220714_MOD-ARA_2021-22.pdf.

Lightning: How the Chinese Military Is Adopting Artificial 71 UK National Audit Office, ‘The Digital Strategy for Defence:

Intelligence’, Center for Security and Emerging Technology, A Review of Early Implementation’, p. 29.

October 2021, p. 10, https://cset.georgetown.edu/publication/ 72 Ibid, p. 32.

harnessed-lightning/. 73 France24, ‘France to Invest €1.5 Billion in Artificial

61 Acharya and Arnold, ‘Chinese Public AI R&D Spending: Intelligence by 2022’, 29 March 2018, https://www.france24.

Provisional Findings’, p. 2. com/en/20180329-france-invest-15-billion-euros-artificial-

62 IISS workshop, ‘Future-proofing Defence: Opportunities and intelligence-AI-technology-2022.

Challenges of Software-defined Defence’. 74 See Ministère des Armées, ‘Projet de Loi de Finances 2023 : Loi

63 Fedasiuk, Melot and Murphy, ‘Harnessed Lightning: How the de Programmation Militaire 2019–2025’, 2023, p. 28, https://

Chinese Military Is Adopting Artificial Intelligence’, pp. 7–8 www.defense.gouv.fr/sites/default/files/ministere-armees/

and 10. Projet%20de%20loi%20de%20finances%20-%202023%20

64 HM Government, ‘Global Britain in a Competitive Age: The -%20LPM%20année%205.pdf; Assemblée Nationale, ‘Avis

Integrated Review of Security, Defence, Development and Fait au Nom de la Commission de la Défense Nationale

Foreign Policy’, 16 March 2021, p. 38, https://assets.publishing. et des Forces Armées sur le Projet de la Loi de Finances

service.gov.uk/government/uploads/system/uploads/ pour 2022 (n° 4482), Tome II, Défense, Environnement

attachment_data/file/975077/Global_Britain_in_a_Competitive_ et Prospective de la Politique de Défense’, par M. Fabien

Age-_the_Integrated_Review_of_Security__Defence__ Gouttefarde, Député, 20 October 2021, https://www.

Development_and_Foreign_Policy.pdf; see also Ministry of assemblee-nationale.fr/dyn/15/rapports/cion_def/l15b4601-

Defence, ‘Ministry of Defence’s Science and Technology Portfolio’, tii_rapport-avis#_Toc256000032; and Agence d’Innovation

of Defence’s Science and Technology portfolio,’ updated 23 de Défense, ‘Innovation de Défense: Bilan d’Activités 2021’,

45 The International Institute for Strategic Studies

 2022, https://www.defense.gouv.fr/sites/default/files/aid/ 86 Lisa Daigle, ‘German Companies Team for Digitization

Bilan%20d%27activités%202021.pdf. of Germany’s Ground Forces’, 27 March 2017, https://

75 Vivienne Machi, ‘France Approves Final Phase of Artemis Big- militaryembedded.com/comms/communications/german-

data Processing Platform’, DefenseNews, 11 July 2022, https:// companies-team-for-digitization-of-germanys-ground-forces.

www.defensenews.com/global/europe/2022/07/11/france- 87 See, for example, Thales Group, ‘D-LBO (Digital Landbased

approves-final-phase-of-artemis-big-data-processing-platform/. Operations)’ and ‘TEN (Tactical Edge Networking)’, 2021,

76 Assemblée Nationale, ‘Avis Fait au Nom de la Commission https://www.thalesgroup.com/en/europe/germany/d-lbo-

de la Défense Nationale et des Forces Armées sur le Projet ten; and Systematic, ‘C4I Software for the Digitalisation

de la Loi de Finances pour 2022 (n° 4482), Tome II, Défense, of the German Army’, 10 March 2022, https://systematic.

Environnement et Prospective de la Politique de Défense’, com/en-gb/industries/defence/news-knowledge/

par M. Fabien Gouttefarde, Député. news/2022_c4i-across-the-german-armed-forces/.

77 Ibid. 88 Peter Hille and Nina Werkhäuser, ‘The German military’s new

78 Ibid. shopping list’, DW, 6 March 2022, https://www.dw.com/en/

79 Laurent Lagneau, ‘Renseignement: Le ministère des Armées how-will-the-german-military-spend-100-billion/a-62020972.

lance le projet TORNADE pour le traitement massif des 89 US Air Force, ‘AFVentures: FY18-FY20 Impact Report’,

données’, Opex360.com, 12 October 2022, https://www.opex360. 2021, p. 2, https://afwerx.com/wp-content/uploads/2021/10/

com/2022/10/12/renseignement-le-ministere-des-armees-lance- AFVentures-2020-Annual-Report.pdf.

le-projet-tornade-pour-le-traitement-massif-des-donnees/. 90 Simona R. Soare, ‘European Military AI: Why Regional

80 Vivienne Machi, ‘French 2023 defense budget adds $3 billion to Approaches Are Lagging Behind’, in Michael Raska and

fund “war economy”’, DefenseNews, 28 September 2022, https:// Richard Bitzinger (eds.), Global Strategic Perspectives on

www.defensenews.com/global/europe/2022/09/28/french-2023- Military AI (London: Routledge, forthcoming 2023).

defense-budget-adds-3-billion-to-fund-war-economy/. 91 Assemblée Nationale, ‘Avis fait au nom de la Commission

81 Agence d’Innovation de Défense, ‘Innovation de Défense de la Défense Nationale et des Forces Armées sur le Projet

Bilan d’Activités 2021’, p. 12. de la Loi de finances pour 2022 (n° 4482), Tome II, Défense,

82 Lagneau, ‘Renseignement: Le ministère des Armées lance le Environnement et Prospective de la Politique de Défense’,

projet TORNADE pour le traitement massif des données’. See par M. Fabien Gouttefarde, Député.

also Earthcube, ‘Press Release: Series-A and Name Change’, 92 Ibid and Agence d’Innovation de Défense, ‘Innovation de

19 November 2020, https://www.preligens.com/resources/ Défense: Bilan d’Activités 2021’.

press/press-release-series-name-change. 93 Bpifrance, ‘Definvest, le Fonds du Ministère des Armées

83 Bastian Giegerich and Ben Schreer, ‘Germany’s New Defence géré par Bpifrance, soutient le développement de pme

Policy: The 100 Billion Euro Question’, IISS Military Balance Blog, 1 stratégiques de la défense’, Fusions et Acquisitions Magazine,

March2022,https://www.iiss.org/blogs/military-balance/2022/03/ January–February 2019, p. 4, https://fusions-acquisitions.info/

germanys-new-defence-policy-the-100-billion-euro-question. wp-content/uploads/2020/05/FA-301-3-5-1.pdf.
84 German Ministry of Defence, ‘Defence Budget 2021’, December 94 Authors’ calculations based on data on overall budgets and

2020, https://www.bmvg.de/de/themen/verteidigungshaushalt/ new competitions in several US defence agencies, including

verteidigungshaushalt-2021. See also Christian Mölling and Torben the Defence Advanced Research Projects Agency (DARPA),

Schütz, ‘Defence Innovation: New Models and Procurement Defence Information Systems Agency (DISA), Defence

Implications: The German Case’, Armament Industry European Security Cooperation Agency (DSCA), Defence Threat

Research Group, May 2021, pp. 6–7, https://www.iris-france.org/ Reduction Agency (DTRA), Missile Defence Agency (MDA),

wp-content/uploads/2021/05/68-Policy-Paper-Def-Innov-German- US Special Operations Forces (SOF) and National Science

Case-May-2021.pdf. Foundation (NSF), in FY 21.

85 German Federal Ministry of Finance, ‘Discover the Federal Budget 95 Authors’ calculations based on data on new competitions

Interactively: Federal Ministry of Defence 2023’, December 2022, launched by the Dstl and DASA in the UK, the AID in France

https://www.bundeshaushalt.de/DE/Bundeshaushalt-digital/ and the Cyber Innovation Hub in Germany between 1 June

bundeshaushalt-digital.html. 2021–31 December 2022.

Software-defined Defence: Algorithms at War 46

96 Simona R. Soare, ‘Cooperative edge: Key Drivers of Defence Defence Analyses, October 2019, pp. 7–8, https://www.ida.

Innovation in Multinational Institutions like NATO and the org/-/media/feature/publications/f/fa/factors-limiting-the-

EU’, IISS, forthcoming 2023. speed-of-software-acquisition/d-10907.ashx.

97 Fields et al., ‘Design and Acquisition of Software for Defense 103 US Government Accountability Office, ‘Weapon Systems

Systems’, p. 3 and Patterson, ‘From Software to Battlespace: Annual Assessment: Challenges to Fielding Capabilities

Reimagining how Defence does DevSecOps with D2S’. Faster Persist’, GAO-22-105230, June 2022, p. 129, https://

98 US Government Accountability Office, ‘DOD Software www.gao.gov/assets/gao-22-105230.pdf.

Acquisition: Status of and Challenges Related to Reform 104 Author calculations based on data from Government

Efforts’, GAO-21-105298. Accountability Office, ‘Report to Congressional Committees:

99 Fields et al., ‘Design and Acquisition of Software for Defense Weapon Systems Annual Assessment. Challenges to Fielding

Systems’, pp. 13 and 18. Capabilities Faster Persist’.

100 German Federal Ministry of Defence, ‘Information from the 105 Ibid, p. 93.

Parliamentary Commissioner for the Armed Forces: Annual 106 Ibid.

Report 2020 (62nd Report)’, 23 February 2021, p. 94, https://www. 107 Ibid.

bundestag.de/resource/blob/839328/e1a864120697c2705 108 Comprehensive analysis of defence software development

7534944ceb20111/annual_report_2020_62nd_report-data.pdf. shortcomings in the US can be found in Office of the

101 The US DoD’s taxonomy on software distinguishes between Undersecretary of Defense for Research and Engineering,

enterprise systems, business systems and combat systems. ‘Design and Acquisition of Software for Defense Systems’, US

Enterprise systems are very large-scale software systems used Defense Science Board, 14 February 2018, https://apps.dtic.mil/

at the DoD level for a variety of purposes and which rely on sti/pdfs/AD1048883.pdf; US Government Accountability Office,

unmodified commercial off-the-shelf software, but with a DoD- ‘Artificial Intelligence: Status of Developing and Acquiring

specific configuration. Notable examples are email systems, Capabilities for Weapon Systems’, GAO-22-10476, February

accounting systems, travel systems and human-resources 2022, https://www.gao.gov/assets/gao-22-104765.pdf; US

databases. Business systems are also very large-scale software Government Accountability Office, ‘DOD Software Acquisition:

systems which generally operate at agency or service level. Status of and Challenges Related to Reform Efforts’, GAO-21-

Examples include software-development environments and 105298; and Kevin Garrison, David M. Tate and John W. Bailey,

logistics systems. Combat systems are unique to defence and ‘Factors Limiting the Speed of Software Acquisition’, Institute

often require at least some level of customisation. Combat for Defense Analyses, D-10907, October 2019, https://www.

systems include logistics systems (software to track materials, ida.org/-/media/feature/publications/f/fa/factors-limiting-the-

supplies and spare parts, transport as part of operational use), speed-of-software-acquisition/d-10907.ashx.

mission systems (mission planning and monitoring software 109 US Department of Defense, ‘Summary of the 2018 National

systems, including modelling and simulation) and weapons Defence Strategy of the United States of America: Sharpening
systems (any software which directly engages or supports lethal the American Military’s Competitive Edge’, 22 January
force and the operation of a weapon platform). To enable greater 2018, p. 10, https://dod.defense.gov/Portals/1/Documents/
warfighting advantage, mission and weapons systems require pubs/2018-National-Defense-Strategy-Summary.pdf.
new software-defined functionality frequently (approximated 110 White House, ‘National Security Strategy’, 12 October 2022, p.

in days to months). However, weapon-systems software is the 22, https://www.whitehouse.gov/wp-content/uploads/2022/10/

most challenging to update regularly, as it is traditionally closely Biden-Harris-Administrations-National-Security-

tied to hardware, one of the enduring challenges of agile and Strategy-10.2022.pdf.

iterative approaches to software development. See US Defence 111 US Department of Defense, ‘2022 National Defense Strategy

Innovation Board, ‘Software Acquisition and Practices (SWAP) of the United States of America, Including the 2022 Nuclear

Study’, May 2019, pp. 2–3, https://media.defense.gov/2019/ Posture Review and the 2022 Missile Defence Review’,

May/01/2002126693/-1/-1/0/SWAP%20MAIN%20REPORT.PDF. 27 October 2022, p. 19, https://media.defense.gov/2022/

102 Kevin Garrison, David M. Tate and John W. Bailey, ‘Factors Oct/27/2003103845/-1/-1/1/2022-NATIONAL-DEFENSE-

Limiting the Speed of Software Acquisition’, Institute for STRATEGY-NPR-MDR.PDF.

47 The International Institute for Strategic Studies

112 US Department of Defense, ‘2019 Digital Modernization 四个五年规划和2035年远景目标纲要 [Outline of the 14th Five-

Strategy’, 12 July 2019, p. 44, https://media.defense. Year Plan for National Economic and Social Development of

gov/2019/Jul/12/2002156622/-1/-1/1/DOD-DIGITAL- the People’s Republic of China and the Long-Range Goals for

MODERNIZATION-STRATEGY-2019.PDF. 2035], 13 March 2021, http://www.gov.cn/xinwen/2021-03/13/

113 US Department of Defense, ‘Executive Summary: DoD Data content_5592681.htm.

Strategy’, 30 September 2020, https://media.defense.gov/2020/ 126 The State Council of the People’s Republic of China,

Oct/08/2002514180/-1/-1/0/DOD-DATA-STRATEGY.PDF. ‘Guowuyuan guanyu yinfa xin yidai rengong zhineng

114 US Deputy Secretary of Defense, ‘Deputy Secretary of fazhan guihua de tongzhi guo fa’ 国务院关于印发新一代人

Defense Memorandum for Senior Pentagon Leadership, 工智能发展规划的通知国发 [The State Council on Issuance of

Commanders of the Combatant Commands, Defense Agency the Development Plan for the New Generation of Artificial

and DoD Field Activity Directors, Subject: Initial Operating Intelligence], No. 35, 20 July 2017, http://www.gov.cn/

Capability of the Chief Digital and Artificial Intelligence zhengce/content/2017-07/20/content_5211996.htm.

Officer’, 1 February 2022, https://media.defense.gov/2022/ 127 Li Changhai 李长海, Han Jian 韩剑 and Zhu Hao 朱昊, ‘Yong

Feb/02/2002931807/-1/-1/1/MEMORANDUM-ON-THE- shuju shuohua de shidai, da shuju ruhe tuijin zhuangbei

INITIAL-OPERATING-CAPABILITY-OF-THE-CHIEF- baozhang biange’ 用数据说话的时代，大数据如何推进装备保

DIGITAL-AND-ARTIFICIAL-INTELLIGENCE-OFFICER.PDF. 障变革 [In the Era of Using Data, How Can Big Data Advance

115 Govini, ‘The 2020 Federal Scorecard’, 2020, p. 27, https:// Equipment Security Changes], PLA Daily, 7 July 2020, http://

govini.com/research/the-2020-federal-scorecard/. www.81.cn/jmywyl/2020-07/07/content_9847982.htm.

116 Ibid. 128 Song Yuangang 宋元刚, Shao Longfei 邵龙飞 and Wang Han

117 Ibid. 特约记者 王晗, ‘Junshi da shuju: Junshi zhineng biange de

118 US Department of Defense, ‘Hicks Announces New Artificial jiasuqi’ 军事大数据：军事智能变革的加速器 [Military Big

Intelligence Initiative’, DoD News, 22 June 2021, https://www. Data: An Accelerator of Change for Military Intelligence],

defense.gov/News/News-Stories/Article/Article/2667212/ PLA Daily, 6 September 2019, http://www.81.cn/bqtd/2019-

hicks-announces-new-artificial-intelligence-initiative/. 09/06/content_9632838.htm.

119 US Government Accountability Office, ‘Artificial Intelligence: 129 The State Council of the People’s Republic of China, ‘Xi Jinping:

Status of Developing and Acquiring Capabilities for Weapons Gaoju zhongguo tese shehui zhuyi weida qizhi wei quanmian

Systems’, GAO-22-104765. jianshe shehui zhuyi xiandaihua guojia er tuanjie fendou—zai

120 New York University, ‘DARPA Award Aims For Autonomous zhongguo gongchandang de ershi ci quanguo daibiao dahui

Teams Of Robots’, Science Blog, 29 September 2022, https:// shang de baogao’ 习近平：高举中国特色社会主义伟大旗帜 为
scienceblog.com/534015/darpa-award-aims-for-autonomous- 全面建设社会主义现代化国家而团结奋斗——在中国共产党第

teams-of-robots/. 二十次全国代表大会上的报告 [Xi Jinping: Holding High the

121 US Department of Defense, ‘2019 Digital Modernization Great Banner of Socialism with Chinese Characteristics and

Strategy’. Uniting the Struggle for the Comprehensive Construction

122 National Security Commission on Artificial Intelligence, of a Modern Socialist Country - Report at the 20th National
‘Final Report’, 19 March 2021, https://www.nscai.gov/ Congress of the Communist Party of China], 25 October 2022,
wp-content/uploads/2021/03/Full-Report-Digital-1.pdf. http://www.gov.cn/zhuanti/zggcddescqgdbdh/sybgqw.htm.
123 US Congress, ‘Statement of Dr. Eric Schmidt to the House Armed 130 The State Council of the People’s Republic of China, ‘National

Services Committee’, 17 April 2018, https://es.ndu.edu/Portals/75/ New Generation AI Plan’, The OECD Artificial Intelligence

Documents/HHRG-115-AS00-Wstate-SchmidtE-20180417.pdf. Policy Observatory, 5 September 2022, https://oecd.ai/

124 Author interview with the Department of Defense’s Chief en/dashboards/policy-initiatives/http:%2F%2Faipo.oecd.

Digital and AI Officer Craig Martell, 21 October 2022. org%2F2021-data-policyInitiatives-24274, https://oecd.

125 The State Council of the People’s Republic of China, ai/en/dashboards/policy-initiatives/http://aipo.oecd.

‘Zhonghua renmin gongheguo guomin jingji he shehui org/2021-data-policyInitiatives-24274.

fazhan di shisi ge wu nian guihua he 2035 nian yuanjing 131 Acharya and Arnold, ‘Chinese Public AI R&D Spending:

mubiao gangyao’ 中华人民共和国国民经济和社会发展第十 Provisional Findings’.

Software-defined Defence: Algorithms at War 48

132 Yuangang, Longfei and Han, ‘Junshi da shuju: Junshi zhineng 141 Andrea Gilli and Mauro Gilli, ‘Why China Has Not Caught

biange de jiasuqi’ 军事大数据：军事智能变革的加速器 [Military Up Yet: Military-Technological Superiority and the Limits

Big Data: An Accelerator of Change for Military Intelligence]. of Imitation, Reverse Engineering, and Cyber Espionage’,

133 FBI, ‘Chinese Military Hackers Charged in Equifax Breach’, International Security, vol. 43, no. 3, Winter 2018/2019, pp. 141–189.

10 February 2020, https://www.fbi.gov/news/stories/ 142 Elsa B. Kania, ‘“AI Weapons” in China’s Military Innovation’,

chinese-hackers-charged-in-equifax-breach-021020. Brookings Institution, Global China: Assessing China’s Growing Role

134 Matt Pottinger and David Feith, ‘The Most Powerful Data in the World, April 2020, https://www.brookings.edu/wp-content/

Broker in the World Is Winning the War Against the U.S.’, uploads/2020/04/FP_20200427_ai_weapons_kania_v2.pdf.

New York Times opinion piece, 30 November 2021, https:// 143 Bureau of Industry and Security, ‘Implementation of

www.nytimes.com/2021/11/30/opinion/xi-jinping-china-us- Additional Export Controls: Certain Advanced Computing

data-war.html. and Semiconductor Manufacturing Items; Supercomputer

135 Wu Min 吴敏, Chen Fengjun 陈凤军 and Zhang Dan 张丹, and Semiconductor End Use; Entity List Modification’,

‘Shenhua da shuju jishu zai junshi lingyu de yingyong’ 深 Federal Register, vol. 87, no. 197, 13 October 2022, https://

化大数据技术在军事领域的应用 [Deepen the Application www.federalregister.gov/documents/2022/10/13/2022-21658/

of Big Data Technology in the Military Field], PLA Daily, 2 implementation-of-additional-export-controls-certain-

March 2021, http://www.81.cn/jfjbmap/content/2021-03/02/ advanced-computing-and-semiconductor; and Nigel

content_283870.htm. Inkster, Emily S. Weinstein and John Lee, ‘Ask the Experts:

136 School of Shipbuilding, ‘Zhongguo shou ge shi haikuang Is China’s Semiconductor Strategy Working?’, LSE Blogs,

zhineng chuan ting jingsai zai shanghai jiaoda haiyang 1 September 2022, https://blogs.lse.ac.uk/cff/2022/09/01/

zhuangbei zhineng yanjin zhongxin juban’ 中国首个实 is-chinas-semiconductor-strategy-working/.

海况智能船艇竞赛在上海交大海洋装备智能演进中心举办 144 Mathieu Duchâtel, ‘The Weak Links in China’s Drive for

[China’s first real-sea intelligent boat competition was held at Semiconductors’, Institute Montaigne Policy Paper, January
Shanghai Jiaotong University Marine Equipment Intelligent 2021, https://www.institutmontaigne.org/en/publications/
Evolution Centre], Shanghai Jiaotong University, 15 October weak-links-chinas-drive-semiconductors.
2019, https://news.sjtu.edu.cn/jdyw/20191014/112627.html. 145 Che Pan, ‘Tech War: Beijing, Local Governments Pressed to

137 Cyberspace Administration of China, ‘Yi zhi zhangyu Raise Support for Chinese Chip Research and Development

gaibianle wangluo anquan youxi guize’ 只章鱼改变了网 Amid US Expansion’, South China Morning Post, 26 August

络安全游戏规则 [An Octopus Changed the Rules of the 2022, https://www.scmp.com/tech/tech-war/article/3190323/

Cybersecurity Game], Science and Technology Daily, 28 May tech-war-beijing-local-governments-pressed-raise-support-

2019, http://www.cac.gov.cn/2019-05/28/c_1124549858.htm. chinese-chip.

138 Yang Wei, ‘Development of future fighters’, Acta Aeronautica 146 Science and Technology News, ‘Nvidia Makes New

et Astronautica Sinica, vol. 41, no. 6, June 2020, pp. 8–19. “Advanced” AI Chip for China That Meets Trade Restrictions’,
139 Ministry of Foreign Affairs of the People’s Republic of 8 November 2022, https://www.technology.org/2022/11/08/
China, ‘Zhongguo guanyu guifan rengong zhineng junshi nvidia-makes-new-advanced-ai-chip-for-china-that-meets-
yingyong de lichang wenjian’ 中国关于规范人工智能军事应 trade-restrictions/.
用的立场文件 [China’s Position Paper on Regulating Military 147 Inkster, Weinstein and Lee, ‘Ask the Experts: Is China’s

Applications of Artificial Intelligence], 14 December 2021, Semiconductor Strategy Working?’.

https://www.fmprc.gov.cn/web/wjb_673085/zfxxgk_674865/ 148 Pan, ‘Tech War: Beijing, Local Governments Pressed to Raise

gknrlb/tywj/zcwj/202112/t20211214_10469511.shtml. Support for Chinese Chip Research and Development Amid

140 Tai Ming Cheung, ‘Strengths and Weaknesses of China’s US Expansion’.

Defense Industry and Acquisition System and Implications 149 For an in-depth analysis of the relationship between national

for the United States’, Naval Postgraduate School, defence AI/ML capability development and multinational

Acquisition Research Program Sponsored Report Series, 25 initiatives within NATO and the EU see Simona R. Soare,

June 2018, https://dair.nps.edu/bitstream/123456789/2724/1/ ‘European Military AI: Why Regional Approaches are

UCSD-AM-18-218.pdf. Lagging Behind’.

49 The International Institute for Strategic Studies

150 Author interview with former French defence official, 2022. https://www.gov.uk/government/news/mod-awards-38-

151 Tom Copinger-Symes, ‘For National Defence, a Digital million-contract-for-advanced-base-protection-system.

Arsenal Is Now Vital’, Wired, 18 February 2022, https://www. 165 Kenneth Payne, ‘Bright Prospects – Big Challenges: Defence

wired.co.uk/article/national-defence-digital-warfare. AI in the United Kingdom’, Defence AI Observatory, 2022, pp.

152 Soare and Pothier, ‘Leading Edge: Key Drivers of Defence 15–16, https://defenseai.eu/wp-content/uploads/2023/01/DAIO_

Innovation and the Future of Operational Advantage’. Study2204.pdf.

153 UK Ministry of Defence, ‘The Defence Capability Framework’, 166 The Stack, ‘UK’s Ministry of Defence Signs £75m Deal with

6 July 2022, https://www.gov.uk/government/publications/ Palantir’, 22 December 2022, https://thestack.technology/

the-defence-capability-framework. uk-ministry-of-defence-palantir-contract-mod/.

154 UK Ministry of Defence, ‘Integrated Operating Concept’, 167 UK Ministry of Defence, ‘MOD Awards £3.8-million Contract

August 2021, p. 7, https://assets.publishing.service.gov.uk/ for Advanced Base Protection System’.

government/uploads/system/uploads/attachment_data/ 168 Royal Navy, ‘New Testbed Ship to Enhance Experimentation

file/1014659/Integrated_Operating_Concept_2025.pdf. in Royal Navy’, 29 July 2022, https://www.royalnavy.mod.uk/

155 UK Ministry of Defence, ‘Data Strategy for Defence: news-and-latest-activity/news/2022/july/29/20220729-new-

Delivering the Defence Data Framework and Exploiting the testbed-ship-to-enhance-experimentation-in-royal-navy.

Power of Data’, 27 September 2021, p. 11, https://www.gov. 169 Olivia Savage, ‘DTDT 2022: Royal Navy Official Calls for New

uk/government/publications/data-strategy-for-defence. Experimentation Facility’, Janes Defence Weekly, 30 September

156 UK Ministry of Defence, ‘Digital Strategy for Defence: 2022, https://www.janes.com/defence-news/news-detail/dtdt-

Delivering the Digital Backbone and Unleashing the Power 2022-royal-navy-official-calls-for-new-experimentation-facility.

of Defence’s Data’, 27 May 2021, p. 10, https://www.gov. 170 Royal Navy, ‘Royal Marines Take Big Step With New

uk/government/publications/digital-strategy-for-defence- Technology’, 25 April 2019, https://www.royalnavy.mod.uk/

delivering-the-digital-backbone-and-unleashing-the-power- news-and-latest-activity/news/2019/april/25/190425-
of-defences-data. commando-warrior.

157 UK Ministry of Defence, ‘Ambitious, Safe, Responsible: 171 Patterson, ‘From Software to Battlespace: Reimagining how

Our Approach to the Delivery of AI-enabled Capability in Defence does DevSecOps with D2S’.

Defence’, June 2022, p. 1, https://assets.publishing.service.gov. 172 Author interview with defence official, July 2022.

uk/government/uploads/system/uploads/attachment_data/ 173 UK Army, ‘Project Convergence 22: What is Sensor-Decider-

file/1082991/20220614-Ambitious_Safe_and_Responsible.pdf. Effector?’, 31 October 2022, https://www.army.mod.uk/

158 UK Ministry of Defence, ‘Defence Artificial Intelligence news-and-events/news/2022/10/project-convergence-

Strategy’, June 2022, https://assets.publishing.service.gov. 22-sensor-decider-effector/.

uk/government/uploads/system/uploads/attachment_data/ 174 Royal Navy, ‘New London Innovation Hub Will See Royal Navy

file/1082416/Defence_Artificial_Intelligence_Strategy.pdf. and US Navy Embrace Technology Together’, 14 June 2022,

159 Ibid, p. 1. https://www.royalnavy.mod.uk/news-and-latest-activity/
160 Ibid, p. 2. news/2022/june/14/220614-london-tech-bridge-launch.
161 UK National Audit Office, ‘The Digital Strategy for Defence: 175 Royal Air Force, ‘RAF FP Force Supporting Innovative Ways

A Review of Early Implementation’, pp. 8–10 and 27. To Provide Force Protection’, 18 November 2021, https://

162 IISS, ‘Defence Innovation Talks: a Conversation with General www.raf.mod.uk/our-organisation/force-protection/news/

Sir Nick Carter’, 30 March 2021, https://www.iiss.org/events/ raf-fp-force-supporting-innovative-ways-to-provide-force-

2021/03/defence-innovation-talks-general-sir-nick-carter. protection/.

163 UK Ministry of Defence, ‘Joint Doctrine Note 1/23: Intelligence, 176 Jamie Grierson, ‘MoD Delivery of Ajax Armoured Vehicles

Surveillance and Reconnaissance’, January 2023, pp. 65–69, https:// will be a Challenge, says Watchdog’, Guardian, 11 March

assets.publishing.service.gov.uk/government/uploads/system/ 2022, https://www.theguardian.com/uk-news/2022/

uploads/attachment_data/file/1130778/JDN_1_23_ISR_web.pdf. m a r / 1 1 / m o d - d e l i ve r y - o f - a j a x - a r m o u r e d - ve h i c l e s -

164 UK Ministry of Defence, ‘MOD Awards £3.8-million Contract will-be-a-challenge-says-watchdog; and UK National

for Advanced Base Protection System’, 10 September 2021, Audit Office, ‘The Ajax Programme’, 11 March 2022,

Software-defined Defence: Algorithms at War 50

 https://www.nao.org.uk/wp-content/uploads/2022/03/The- 191 Assemblée Nationale, ‘Avis fait au nom de la Commission

Ajax-programme.pdf. de la Défense Nationale et des Forces Armées sur le Projet

177 NASA, ‘Storm Clouds over Stonehenge: UK Watchkeeper de la Loi de Finances pour 2022 (n° 4482), Tome II, Défense,

UAS Mishap’, NASA Safety Center, System Failure Case Environnement et Prospective de la Politique de Défense par

Study, vol 19, issue 1, April 2019, https://www.scribd.com/ M. Fabien Gouttefarde, MP’.

document/485908721/Storm-Clouds-Over-Stonehenge#. 192 Marc Watin-Augouard, ‘The French “Digital Defence Agency”

178 UK National Audit Office, ‘The Equipment Plan 2022 to has been established’, inCyber, 25 May 2021, https://incyber.

2032’, p. 8. org/en/the-french-digital-defence-agency-has-been-established/.

179 Author interview with defence representatives, September 2022. 193 Stephanie Pezard, Michael Shurkin and David A. Ochmanek,

180 Ministère des Armées, ‘Artificial Intelligence in Support of ‘A Strong Ally Stretched Thin: An Overview of France’s

Defence: Report of the AI Task Force’, September 2019, p. 5, Defense Capabilities from a Burdensharing Perspective’,

https://www.defense.gouv.fr/sites/default/files/aid/Report%20 RAND Corporation, 2021, p. 30, https://www.rand.org/pubs/

of%20the%20AI%20Task%20Force%20September%202019.pdf. research_reports/RRA231-1.html.

181 Ministère des Armées, ‘Strategic Update 2021: Synthesis’, 194 Ministère des Armées, ‘Vision Stratégique Du Chef D’État-

2021, https://s.rfi.fr/media/display/e19540ea-b16e-11eb-b464- Major Des Armées’, p. 10.

005056bff430/210300%20France%20defense%20strategic- 195 Soare, ‘Cooperative Edge: Key Drivers of Defence Innovation

update%202021.pdf. in Multinational Organisations like NATO and the EU’.

182 Ministère des Armées, ‘Vision Stratégique Du Chef D’État- 196 Jean-Pierre Devaux and Gaspard Schnitzler, ‘Defence Innovation:

Major Des Armées’ [Strategic Vision of the Chief of Defence New Models and Procurement Implications, The French Case’,

Staff], October 2021, https://www.defense.gouv.fr/sites/default/ Armament Industry European Research Group, September 2020,

files/ema/211022_EMACOM_VisionStrategiqueCEMA_FR_ p. 6, https://www.iris-france.org/wp-content/uploads/2020/09/63-

Vdef_HQ%20%282%29.pdf. Policy-Paper-Def-Innov-France-September-2020.pdf.

183 Ministère des Armées, ‘Document de Référence de 197 Ibid, pp. 9–10.

l’Orientation de l’Innovation de Défense’ [Reference Document 198 Sylvie Matelly, ‘Defense Innovation and the Future of

on Orientation, Innovation, and Defence], 2022, https://www. Transatlantic Strategic Superiority: A French Perspective’,

defense.gouv.fr/sites/default/files/aid/DrOID-2022.pdf. German Marshall Fund of the United States, 9 April 2018,

184 Palais Elysée, ‘National Strategic Review 2022’, 2 December https://www.gmfus.org/news/defense-innovation-and-

2022, p. 47, http://www.sgdsn.gouv.fr/uploads/2022/12/rns-uk- future-transatlantic-strategic-superiority-french-perspective.

20221202.pdf. 199 See Thierry Burkhard, ‘French Army Chief: “Military Force

185 Ibid, p. 11. is Making a Brutal Return to the International Scene”’, Le

186 Palais Elysée, ‘Defence and National Security Strategic Review Monde, 11 November 2022, https://www.lemonde.fr/en/

2017’, 15 October 2017, pp. 48 and 68, https://www.dsn.gob.es/ opinion/article/2022/11/11/french-army-chief-military-

sites/dsn/files/2017%20France%20Strategic%20Review.pdf. force-is-making-a-brutal-return-to-the-international-
187 See, for example, the priorities under Ministère des Armées, scene_6003869_23.html; Davide Basso, ‘France Not Ready
‘Document de Référence de l’Orientation de l’Innovation de for High-intensity War Says Former Army Chief’, Euractiv,
Défense’, https://www.defense.gouv.fr/sites/default/files/aid/ 9 November 2022, https://www.euractiv.com/section/
DrOID-2022.pdf. politics/news/france-not-ready-for-high-intensity-war-
188 Ministère des Armées, ‘INSTRUCTION N° 2067/ARM/CAB/ says-former-army-chief/; and Cédric Pietralunga, ‘French
CC6 relative à l’innovation de défense au sein du ministère des Military Chiefs Sound the Alarm on the State of the Armed
Armées’, Cabinet de la Ministre, 7 May 2020, p. 3, https://www. Forces’, Le Monde, 13 August 2022, https://www.lemonde.
defense.gouv.fr/sites/default/files/aid/IMID du 7 mai 2020.pdf. fr/en/international/article/2022/08/13/french-military-
189 Agence d’Innovation de Défense, ‘Innovation de Défense: chiefs-of-staff-sound-the-alert-on-the-state-of-their-
Bilan d’Activitiés 2021’, 2021, p. 14, tps://www.defense.gouv. troops_5993519_4.html.
fr/sites/default/files/aid/Bilan%20d%27activités%202021.pdf. 200 Assemblée Nationale, ‘Avis Fait au Nom de la Commission

190 Ibid. de la Défense Nationale et des Forces Armées sur le Projet

51 The International Institute for Strategic Studies

 de la Loi de Finances pour 2022 (n° 4482), Tome II, Défense, 207 French Land Army, ‘Operational Superiority: A Battle-

Environnement et Prospective de la Politique de Défense par hardened Army Able to Commit in All Conflicts Up to a

M. Fabien Gouttefarde, MP’. Major Confrontation’, 2022, https://www.defense.gouv.fr/

201 Preligens, ‘Earthcube Overshoots Its Competitors’, 22 sites/default/files/terre/2022-Infographie-Vision-Strategique-

September 2020, https://www.preligens.com/resources/ anglaise.pdf.

press/earthcube-overshoots-its-competitors; and Pierre 208 Sopra Steria, ‘Sopra Steria Chosen by the French Ministry of

Tran, ‘French Intelligence Agency Wants AI to Help Sort Defence to Implement Brasidas, the Single New Information

Masses of Raw Data’, DefenseNews, 5 February 2018, System for Aerospace in Service Support’, 25 March 2021,

https://www.defensenews.com/global/europe/2018/02/05/ https://www.soprasteria.com/newsroom/press-releases/

french-intelligence-agency-seeks-ai-to-support-analysts/. details/sopra-steria-chosen-by-the-french-ministry-of-

202 European Defence Review, ‘TALIOS Optronic Pod Qualified defence-to-implement-brasidas-the-single-new-information-

by French Defence Procurement Agency’, EDR Magazine, system-for-aerospace-in-service-support.

19 November 2018, https://www.edrmagazine.eu/talios- 209 Assemblée Nationale, ‘Avis Fait au Nom de la Commission

optronic-pod-qualified-by-french-defence-procurement- de la Défense Nationale et des Forces Armées sur le Projet

agency and MBDA Missile Systems, ‘MBDA Collaboration de la Loi de Finances pour 2022 (n° 4482), Tome II, Défense,

Wins National Engineering Award for Work with Artificial Environnement et Prospective de la Politique de Défense par

Intelligence’, 5 July 2019, https://www.mbda-systems. M. Fabien Gouttefarde, MP’.

com/press-releases/le-programme-2aci-recoit-le-prix-aat- 210 Ministère des Armées, ‘Projet de Loi de Finances 2023 - Loi de

ingenieur-general-chanson/. Programmation Militaire 2019-2025’, p. 28.

203 Thales, ‘Thales and Dassault Aviation Win Contract for 211 Ursula von der Leyen, ‘The Digital Transformation

France’s New Strategic Airborne Intelligence Programme’, 14 of the Bundeswehr’, St. Gallen Business Review, 9

January 2020, https://www.thalesgroup.com/en/worldwide- August 2018, https://www.stgallenbusinessreview.com/

defence/radio-communications/news/thales-and-dassault- the-digital-transformation-of-the-bundeswehr/.

aviation-win-contract-frances-new; Thales, ‘Collaborative 212 Ibid.

Anti-Submarine Warfare’, 2021, https://www.thalesgroup. 213 German Federal Government, ‘Artificial Intelligence Strategy

com/en/markets/defence-and-security/naval-forces/ of the German Federal Government: 2020 Update’, December

underwater-warfare/collaborative-anti-submarine-warfare. 2020, https://www.ki-strategie-deutschland.de/files/downloads/

204 Maggie Gray and Amy Ertan, ‘Artificial Intelligence and Fortschreibung_KI-Strategie_engl.pdf.

Autonomy in the Military: An Overview of NATO Member 214 German Federal Government, ‘Data Strategy of the Federal

States’ Strategies and Deployment. Appendix A – Country Government: An innovation strategy for social progress and

Profiles’, NATO Cooperative Cyber Defence Cyber of Excellence, sustainable growth’, 27 January 2021, https://www.

December 2021, pp. 16–18, https://ccdcoe.org/uploads/2021/12/ bundesregierung.de/breg-de/suche/data-strategy-of-the

Strategies_and_Deployment_Appendix-A_A4.pdf. -federal-german-government-1950612.
205 Assemblée Nationale, ‘Avis Fait au Nom de la Commission 215 Author interview with defence representative, September

de la Défense Nationale et des Forces Armées sur le Projet and December 2022.

de la Loi de Finances pour 2022 (n° 4482), Tome II, Défense, 216 Bundeswehr, ‘Dritter Bericht zur Digitalen Transformation des

Environnement et Prospective de la Politique de Défense par M. Geschäftsbereichs des Bundesministeriums der Verteidigung’

Fabien Gouttefarde, MP’. [Third Report on the Digital Transformation of the Federal

206 Machi, ‘France Approves Final Phase of Artemis Big-data Ministry of Defence], February 2021, p. 4, https://www.bmvg.de/

Processing Platform’ and Atos, ‘ATHEA Selected by the French resource/blob/5060250/6f695b7797b03986dd6eabf9946b5a38/

Ministry of the Armed Forces for Last Phase of Project ARTEMIS. download-3-digitalbericht-data.pdf.

IA’, GlobeNewswire, 11 July 2022, https://www.globenewswire. 217 Ibid.

com/en/news-release/2022/07/11/2477276/0/en/ATHEA-selected- 218 Bundeswehr, ‘Digitalisation in the Army’, https://www.

by-the-French-Ministry-of-the-Armed-Forces-for-last-phase-of- bundeswehr.de/en/organization/army/capabilities/digitalisation.

project-ARTEMIS-IA.html. 219 Ibid.

Software-defined Defence: Algorithms at War 52

220 German Federal Government, ‘Strategy Paper of the Federal Geschäftsbereichs des Bundesministeriums der Verteidigung’
Government on Strengthening the Security and Defence [Third Report on the Digital Transformation of the Federal
Industry’, February 2020, p. 3, https://www.bmwk.de/Redaktion/ Ministry of Defence], p. 39.
DE/Downloads/S-T/strategiepapier-staerkung-sicherits-und- 229 Bundeswehr, Zweiter Bericht zur Digitalen Transformation des

verteidigungsindustrie-en.pdf?__blob=publicationFile&v=4. Geschäftsbereichs des Bundesministeriums der Verteidigung’

221 Bundeswehr, ‘Dritter Bericht zur Digitalen Transformation des [Second Report on the Digital Transformation of the Federal

Geschäftsbereichs des Bundesministeriums der Verteidigung’ Ministry of Defence], March 2020, pp. 19–23, https://www.

[Third Report on the Digital Transformation of the Federal bmvg.de/resource/blob/258260/cc60ba7e2570976df105ba

Ministry of Defence], p. 37. f97080fe45/20200312-download-zweiter-bericht-digitale-

222 Ibid, p. 31. transformation-data.pdf.

223 Gray and Ertan, ‘Artificial Intelligence and Autonomy in 230 Mölling and Schütz, ‘Defence Innovation: New Models and

the Military: An Overview of NATO Member States Strategies Procurement Implications: The German Case’, p. 9.

and Deployment. Appendix A – Country Profiles’, pp. 19–21, 231 German Federal Ministry of Defence, ‘Military Scientific Research

https://ccdcoe.org/uploads/2021/12/Strategies_and_ Annual Report 2021: Defence Research for the German Armed

Deployment_Appendix-A_A4.pdf. Forces’, 2022, https://www.bmvg.de/resource/blob/5536306/2b3

224 German Federal Ministry of Defence, ‘Military Scientific d58bd18d845abe6b2e31d8caea4b6/military-scientific-research-

Research Annual Report 2015: Defence Research for the annual-report-2021-data.pdf.

German Armed Forces’, 2015, https://www.bmvg.de/resource/ 232 Soare and Pothier, ‘Leading Edge: Key Drivers of Defence

blob/13614/49cb1a0b29c0d%2092521c7e2f59a3f6b6e/g-03- Innovation and the Future of Operational Advantage’.

download-military-scientific-research-annual-report-2015- 233 Bastian Giegerich and Maximilian Terhalle, The Responsibility

englisch-data.pdf. to Defend: Rethinking Germany’s Strategic Culture, Adelphi 447

225 Cyber Innovation Hub, ‘Innovation projects’, 2022, (Abingdon: Routledge for the IISS, 2021).

h t t p s : / / w w w . c y b e r i n n o va t i o n h u b . d e / i n n o va t i o n / 234 Giegerich and Schreer, ‘Germany’s New Defence Policy: The

innovationsvorhaben. 100 Billion Euro Question’.

226 German Federal Ministry of Defence, ‘Defense budget 235 Mölling and Schütz, ‘Defence Innovation: New Models and

2021’, December 2020, https://www.bmvg.de/de/themen/ Procurement Implications: The German Case’, p. 10.

verteidigungshaushalt/verteidigungshaushalt-2021. 236 German Federal Ministry of Defence, ‘Information from the

227 Systematic, ‘C4I software for the digitalisation of the German Parliamentary Commissioner for the Armed Forces Annual

Army’, 10 March 2022, https://systematic.com/en-gb/industries Report 2021 (63rd Report)’, 15 March 2022, pp. 52–53, https://
/defence/news-knowledge/news/2022_c4i-across-the- www.bundestag.de/resource/blob/901610/fc410cdd8

german-armed-forces/. 93ba69d52b8cb55ed1fb715/annual_report_2021_63rd_report-

228 Bundeswehr, ‘Dritter Bericht zur Digitalen Transformation des data.pdf.

53 The International Institute for Strategic Studies

© The International Institute for Strategic Studies, 2023

Acknowledgements
The authors are grateful to all the governmental, industry
and academic experts who generously donated their time
and insights during various events and interviews organ-
ised to support this analysis. The authors are also grateful
to Dr Bastian Giegerich and Dr Ben Schreer, who reviewed
and provided comments on earlier versions of the report. The
authors wish to thank Beryl Thomas and Fara Breuer for the
research assistance provided during the project.

The International Institute for Strategic Studies – UK

Arundel House | 6 Temple Place | London | wc2r 2pg | UK
t. +44 (0) 20 7379 7676 f. +44 (0) 20 7836 3108 e. iiss@iiss.org www.iiss.org

The International Institute for Strategic Studies – Americas

2121 K Street, NW | Suite 600 | Washington DC 20037 | USA
t. +1 202 659 1490 f. +1 202 659 1499 e. iiss-americas@iiss.org

The International Institute for Strategic Studies – Asia

9 Raffles Place | #49-01 Republic Plaza | Singapore 048619
t. +65 6499 0055 f. +65 6499 0059 e. iiss-asia@iiss.org

The International Institute for Strategic Studies – Europe

Pariser Platz 6A | 10117 Berlin | Germany
t. +49 30 311 99 300 e. iiss-europe@iiss.org

The International Institute for Strategic Studies – Middle East

14th floor, GBcorp Tower | Bahrain Financial Harbour | Manama | Kingdom of Bahrain
t. +973 1718 1155 f. +973 1710 0155 e. iiss-middleeast@iiss.org