You're overthinking it. You're building on your own definition of what a tool is, but courts are likely to find that a person who used an AI with a specific type of prompt were using it as a tool and are responsible for the clearly stated intent behind their use.

Whether that's actually legal in this case I don't know, but I'm pretty sure courts won't conclude "welp, it was the AI, not the user" in a case like this.

Yeah, probably you're right. Though it is an interesting thing to overthink - we have decided that LLMs are not tools for copyright but are tools for accountability. At some point they'll get smart enough to be accountable, and what happens then?