It's a tool. It isn't human, and (currently) is not intelligent. It's a conversational UI on top of a software program.

So, if that "software program" had a traditional button UI, a button said "Create CSAM," and the user pushed it, the program's creator is not culpable at all for providing that functionality?

I think intent comes into play here. Grok was not created to create CSAM, just like photoshop. But both can be used to create it.

I would agree with this if Grok's interface was "put a pixel there, put a line there, now fill this color there" like Photoshop. But it's not. Generative AI is actively assisting users to perform the specific task described and its programming is participating in that task. It's not just generically placing colors on the screen where the user is pointing.

"if I hired a hitman to kill someone, he does, who's culpable? Me? The hitman? Both?"

It's both. Very simple. You can't get around liability by forming a conspiracy [0].

https://en.wikipedia.org/wiki/Criminal_conspiracy

Right, but the makers of the murder weapon aren't culpable.

Or do you think a Microsoft exec should go to jail every time someone uses it to write a death threat?

The hypothetical imagined hiring an intern to do a crime and supposed that this might make liability harder to determine. It doesn't!

An intern is a human, unlike Microsoft Word or an LLM, which are tools/machines/etc.

Automated DDOS-for-hire services are not legal either. They're tools/machines/etc, possibly running more or less autonomously.

https://www.justice.gov/usao-ak/pr/federal-prosecutors-alask...

I think we all know it's illegal to sell illegal services.