When accessing multiple fields of a structured array, numpy fails to verify that all the fields exist. Instead, it will only return existing fields. If none exist, it return a strange object that has a dtype that is empty. Normally, creating an object with an empty dtype is not supposed to be possible (zeros(shape=(5,), dtype=[]) results as expected in TypeError: Empty data-type), so this might actually be a dual bug.

In [216]: A = numpy.empty(dtype=[("A", "f4"), ("B", "f8")], shape=(5,))

In [217]: A[["AA"]]
Out[217]: 
array([(), (), (), (), ()], 
      dtype='{'names':[], 'formats':[], 'offsets':[], 'itemsize':12}')

In [232]: A[["AA", "B"]]
Out[232]: 
array([(0.0,), (0.0,), (0.0,), (0.0,), (0.0,)], 
      dtype=[('B', '<f8')])