Why Anthropic's 'safe' Mythos-class model won't answer questions about cancer

Anthropic's Claude Fable 5 model, part of the 'Mythos-class' series, includes broad safety safeguards that can block or revert to Opus 4.8 when queries about cybersecurity, biology, or chemistry are flagged. The model's restrictions were implemented to prevent potential misuse in sensitive areas, even if they mistakenly block benign requests.