[marcus_holmes]

That's an interesting question, though.

We hold that LLMs are incapable of generating copyrighted images, so it's not just a tool - if it was just a tool then the author would be able to copyright the images. The courts recognise that an LLM is capable of generating things in its own right (which is why they're not copyrightable - copyrights only protect human works).

So it follows that an LLM must be able to create images itself, separate from the human prompter.

Whether that's enough to absolve the human of the crime, though - IANAL, and I suspect it would take the House of Lords to rule on it definitively.

[antonvs]

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.

[marcus_holmes]

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?