I would personally like to have it working that way.
But I also understand that it wouldn't work for people who have the expectation that once a dangerous content is identified and removed from the internet, the models are re-trained immediately
I hope local-first models like Mistral will fix this. If you run it locally other people with their other expectations have little to say about your LLM.
The rule should be "what you can find in Internet search cannot be dangerous."