[roryirvine]

The analogy holds better if you compare the operators of AI services to the operators of gun services.

A shooting range, for example, must take reasonable care to prevent people from harming themselves or others: hearing protection, safety markings, lane dividers, a well-defined target area, backsplash protection, etc.

I'm not entirely against the "just a tool" interpretation of AI, as long as it's used to cover only the model (and maybe the weights) - an inert lump of code, rather than a public-facing inference service.

(I'm generally baffled by American gun culture, so perhaps it is possible there to recklessly operate an unsafe shooting range with impunity, but I doubt it...)