Possible to monkeypatch numpy array slice behavior?

Question

I know monkeypatching is frowned upon (for good reason), but this would only be in a script that nobody else would ever have to deal with. And I'd like to understand Python a bit better:)

Python doesn't throw an error if the bounds of a slice exceed the length of the thing being sliced. For example:

>>> 'abc'[10:20]
''

Which is all fine and good, but what if you'd like to have an error thrown? Specifically, when using Numpy arrays. I would like to have something like

# This will not work! 
import numpy as np

old_getitem = np.array.__getitem__

def new_getitem(self, val):
    if isinstance(val, slice):
        if val.stop >= len(self):
            raise IndexError
    return old_getitem(self, val)

np.array__getitem__ = new_getitem

but this can't work since np.array doesn't have a __getitem__ attribute

>>> np.array.__getitem__
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
AttributeError: 'builtin_function_or_method' object has no attribute '__getitem__'

what I find odd is that the following works:

>>> a = np.array([1, 2, 3])
>>> a.__getitem__(slice(0,2))
array([1, 2])

My first guess would have been that Numpy is using the attribute of a parent class, but (as far as I know) it doesn't have one as a 'builtin_function_or_method'

So I have two questions. Is it possible to do my silly monkeypatch? And why does calling __getitem__ on a Numpy array work when it doesn't have a __getitem__ attribute?