FWIW, Asimov came up with his '3 laws' as a premise in order to be able to write stories about things that didn't yet exist - and human / societal reactions to them. He also came up with 'positronic brains', which seem to have no grounding in our current real world - again as a premise in order to write stories. I won't get into his 'psychohistory' which suggests the inevitability of social change.
A bit closer to home, B.F. Skinner (the rat as a biological robot guy) wrote some fiction about a future society. I remember being a bit gobsmacked when a classmate brought that up as 'proof' of Skinner's hypotheses.
On to AI harming people: It's already been going on. Israeli forces used an AI they call 'Lavender' to identify Gaza residents as 'Hamas operatives' - and this was used (without a human reviewing the evidence) in selection of places to bomb in their initial air strikes - in response to what absolutely was a horrific action by Hamas. I've no issue with them shooting the people who did this thing, but letting an AI pick out people...?