I think the biggest issue with any sort of AI games is the fact most of them are so open ended, the buck ends with your own creativity. When you are the one in control of so much, you quickly get tired when you run out of the story you want to tell.
I have a bit of hope that some sort of proper setup will fix this issue, especially with multiple agents able to pass json to and from the "game" side and the writer side. One of my ideas is an open-ended roguelike card game where the player runs through a bunch of player-created scenarios and builds their deck by taking things from that world to the next, so people can compare how their build influenced their version of the scenario.
I feel like you have missed the point of this. It isn't to completely absolve the user of liability, it's to prove malice instead of incompetence.
If the user claims that they only authorized the bot to review files, but they've warranted the bot to both scan every file and also send emails to outside sources, the competitors in this case, then you now have proof that the user was planning on committing corporate espionage.
To use a more sane version of an example below, if your dog runs outside the house and mauls a child, you are obviously guilty of negligence, but if there's proof of you unleashing the dog and ordering the attack, you're guilty of murder.
I totally disagree with the idea of making raw text-to-image output from closed source models like Midjourney, but I wonder how much level of control would change most people's minds on if an AI image can become copyrightable.
Would an artist who creates a private AI model on their own art be allowed to copyright everything that comes out? What about if it is a small model, based on top of a much larger model with other people's art but copies their personal artstyle very closely (a low rank adapter, or LoRA, in stable diffusion terms)? What if the artist sketches the outline then uses a model that matches the image to that (known as a controlnet), along with the above?
The answer might just be "never", if Chancellor Masters and Scholars of The University of Oxford v. Narendra Publishing house's ruling is upheld, since mathematical formulas can't be copyrighted, and if AI art is reduced to an output of a mathematical formula, then it makes sense to always treat it as copyright-free.
I have a bit of hope that some sort of proper setup will fix this issue, especially with multiple agents able to pass json to and from the "game" side and the writer side. One of my ideas is an open-ended roguelike card game where the player runs through a bunch of player-created scenarios and builds their deck by taking things from that world to the next, so people can compare how their build influenced their version of the scenario.