Anthropic improves Fable 5 biology safeguards while cutting false positives
Anthropic says a retrained biology classifier cut Fable 5 biology fallbacks by about 85% while continuing to reroute harmful and dual-use research requests to Claude Opus 5, widening benign access without intentionally loosening high-risk boundaries.
AWAY FROM DOOM20confidence 66/100
Why it moved the index
A more precise deployed classifier directly strengthens control over frontier biological capability by preserving blocks on dual-use virology, toxicology, and molecular design while reducing benign refusals. Magnitude 20 reflects a narrow safeguard update, and confidence 66 reflects primary but developer-only testing.
AUDIT TRAIL
Assessment history
- R1Away 20 · confidence 66
Initial inclusion from Anthropic's dated deployed-safeguard update with quantified fallback changes and stated retained risk boundaries.
11 Aug 2026