Anthropic updates Fable 5 biology safeguards, cuts false positives by 85%

On August 7, Anthropic published a post announcing updates to Claude Fable 5 biology safeguards that substantially reduce false positives. In Anthropic testing, the update reduced biology-related fallbacks by about 85% across product surfaces. Total fallback volume dropped roughly 67% on Claude.ai and 55% on Cowork. Fable 5 can now assist with a wider range of biology tasks.

Users should see far fewer fallbacks when interpreting lab results, understanding symptoms, and learning about biology in educational contexts. Healthcare professionals will get more support from Fable 5 on clinical tasks. However, requests in virology, toxicology, and molecular design remain dual-use and continue to fall back to Opus 5. Fable 5 is not yet usable for professional biology research and drug development.

Anthropic said Fable 5 can now outperform experts on some highly complex biology tasks and provide operational support on others. That means it can genuinely help a researcher developing a new medical treatment, but it could also be misused by a malicious actor developing a biological weapon. Anthropic committed to investing in responsible ways to give biologists frontier access, including trusted access programs.

The safeguards work via classifiers that detect biology queries or harmful outputs. When a classifier fires, requests route to Opus 5. Anthropic tuned the classifiers to more accurately distinguish benign questions from high-risk ones.