Recent text-to-image models have improved enough such that it's possible to get realistic, not-Midjourney-dreaminess in the generations with a modicium of effort, so banning obviously-AI-generated images is shortsighted and unsustainable.
The whole space is somewhat amusing to me. what is the bigger moral hazard: Openly disclosing everything about your content pipeline and getting your team's efforts shitcanned, or keeping everything private unless a court order shows up?
What would an satisfy an audit trail that no tainted AI data have made it into a digital image? It would involve a chain of attribution per fraction of a pixel through all it's past iterations.
Which wouldn't be sufficient since, as stated many times before, the diffusion process most text-to-image AIs use is not collaging.
This has been the legal street smarts for a while and doesn't seem like a big development to me. As usual, you don't admit or allude to anything. It's like when I'd write code and say no I've never even visited StackOverflow.