Yes, if it was a human agent they would certainly be liable for the mistake, and the law very much already recognises that.

That’s my whole point here; the company should be equally liable for the behaviour of an AI agent as they are for the behaviour of a human agent when it gives plausible but wrong information.