Anthropic improves biology safeguards for Claude Fable 5

Anthropic has announced significant updates to the biology safeguards in its flagship AI model, Claude Fable 5. The company has successfully reduced biology-related fallbacks by 85%, meaning users will encounter fewer instances where the system automatically reverts to a less capable model, such as Opus 5, when handling biological queries.

This improvement aims to balance the model's advanced reasoning capabilities with the need to mitigate dual-use risks, such as the potential exploitation of the AI for creating biological weapons. By refining safety classifiers through rewritten operational rules and updated training data, Anthropic has reduced false positives that previously blocked benign queries.

The update is expected to expand the utility of Claude Fable 5 in health and educational contexts, including assisting with clinical tasks, interpreting lab results, and understanding symptoms, while maintaining strict protections against harmful requests.