Currently

In [1]: a = np.ones(10)[::2]

In [2]: a.flags
Out[2]: 
  C_CONTIGUOUS : False
  F_CONTIGUOUS : False
  OWNDATA : False
  WRITEABLE : True
  ALIGNED : True
  UPDATEIFCOPY : False

In [3]: b = np.array(a.flat, copy=False)

In [4]: a.flags
Out[4]: 
  C_CONTIGUOUS : False
  F_CONTIGUOUS : False
  OWNDATA : False
  WRITEABLE : False
  ALIGNED : True
  UPDATEIFCOPY : False

The reason is that a.flat.__array__ is called which returns a new array with UPDATEIFCOPY and locks a by clearing the WRITEABLE flag. Because b hangs onto a reference of the uncopied array, it does not get deallocated, hence does not re-enable the WRITEABLE flag.

Whether this is a bug or a feature is debatable, but given the usual use of asarray (copy=False), I think that a copy of the array should be made when UPDATEIFCOPY is set.