My guess is that it’s gonna wind up being a split, and it’s not going to be unique to “AI” relative to any other kind of device.
There’s going to be some kind of reasonable expectation for how a device using AI should act, and then if the device acts within those expectations and causes harm, it’s the person who decided to use it.
But if the device doesn’t act within those expectations, then it’s not them, may be the device manufacturer.