As has been made very clear, it is not actually possible to prevent these models from regurgitating any information they’ve been trained on, no matter how fancy the system prompt. So, if there is NSFW content In the training data, users will always be able to access it, not matter how “compliant” the company is with restrictions on NSWF content by way of system prompts.

They can have their cake and eat it to, many users will prefer the models because of their ability to do porn stuff, and they will not be held legally liable for that since they’ve done everything they possible could.

So long as no one proves that they did in fact intentionally train the models on a shit ton of porn …