Anthropic Updates Claude Fable 5 Biology Safeguards, Reduces False Positives by 85%
Anthropic has revised the biology safety measures of its Claude Fable 5 model, achieving an 85% reduction in biology-related fallbacks, thereby improving user experience and accuracy. This enhancement enables Fable 5 to effectively assist with various biology-related tasks, such as interpreting lab results and addressing educational inquiries, while offering increased support to healthcare professionals. Nevertheless, it still depends on Opus 5 for handling dual-use requests in fields like virology, toxicology, and molecular design, rendering it inadequate for professional research. The previous safeguards were extensive to prevent misuse, but the newly updated classifiers now activate less frequently for harmless requests while blocking dangerous content. Anthropic aims to cut total fallbacks by around 67% on Claude.ai and is actively seeking user input.
Key facts
- Anthropic updated Claude Fable 5's biology safeguards, reducing biology-related fallbacks by about 85%.
- Fable 5 can now assist with a wider range of biology tasks, including everyday health and educational questions.
- Healthcare professionals will receive more support from Fable 5 on clinical tasks.
- Fable 5 still falls back to Opus 5 for dual-use requests including virology, toxicology, and molecular design.
- The original safeguards were intentionally broad, blocking almost all biology queries at launch.
- Safety classifiers detect safeguarded biology tasks and reroute requests to Opus 5.
- Anthropic rewrote the classifier's constitution and retrained it with updated data.
- Expected total fallback reductions: 67% on Claude.ai, 55% on Cowork, 17% on Claude Code, 7% on Claude Platform.
Entities
Institutions
- Anthropic
- US Intelligence Community