back

We have no proof that AI models suffer, but Anthropic acts like they might for training purposes.