@user1121588 answered most of this in a comment, but to answer fully…
“Bitwise and” (&) behaves much the same as logical_and on boolean arrays, but it doesn’t convey the intent as well as using logical_and, and raises the possibility of getting misleading answers in non-trivial cases (packed or sparse arrays, maybe).
To use logical_and on multiple arrays, do:
np.logical_and.reduce([a, b, c])
where the argument is a list of as many arrays as you wish to logical_and together. They should all be the same shape.