Anthropic cuts Fable 5's biology fallbacks 85% with a retrained safety classifier
Anthropic rewrote and retrained the classifier behind Fable 5's biology safeguards, cutting biology-related fallbacks about 85% and total fallbacks 67% on Claude.ai, 55% on Cowork, and 17% on Claude Code. Everyday clinical and educational biology questions now reach the frontier model, while virology, toxicology, and molecular design keep routing to Opus 5 — professional biology research stays out of reach until trusted-access pathways exist.
Source: anthropic.com ↗
It's often difficult to tell apart beneficial and harmful uses of AI in biology.
Anthropic
Why this matters
- → Legitimate biologists and clinicians now get Fable 5's full power instead of fallbacks to weaker models
- → Keeps dangerous dual-use research (virology, toxicology, molecular design) gated until trusted access exists
- → Shows how to loosen AI safeguards without opening frontier capabilities to misuse
Safety meets capability