Unfortunately the test cases do not match the description. There shouldn't be any negative input values at all.
That said, it seems, that with the current (haskell) tests it is expected to treat negative values like any other value. Zero is only expected if one of the inputs is zero.

I think this kata needs better tests and a more precise description.

I think your last suggestion could go amiss, but I'll remove the tag and add some example tests.