Anthropic said on August 7, 2026 that it has adjusted the biology safeguards on Fable 5 1. The work centred on the classifier’s constitution — described by the company as “a collection of rules to help the model discern between safeguarded and allowed content” — which Anthropic rewrote before building fresh training data from it and retraining the classifier 1.
What the safeguard actually does is switch models rather than decline. For requests the company treats as dual-use, Fable “still falls back to Opus 5” — a category Anthropic gives as virology, toxicology, and molecular design 1. The change is in where the line sits. Anthropic says users should hit far fewer fallbacks on ordinary health and educational questions, naming interpreting lab results, understanding symptoms, and learning biology in an educational context as examples 1. Healthcare professionals should also get more support on clinical tasks, according to the company 1.
On results, Anthropic reports that in its testing the update cut biology-related fallbacks by about 85% across its product surfaces 1. A separate figure covers fallbacks of every kind, not just biological ones: the company says it expects that total to fall by roughly 67% on Claude.ai, 55% on Cowork, 17% on Claude Code, and 7% on the Claude Platform 1. That second set is an expectation rather than a measured outcome. Both come from Anthropic itself, with no measurement window or sample size given. The company also notes that false positives — cases that carry very little risk but still trip the classifier — will “inevitably remain” 1.
Sources
- Improving Fable 5’s biology safeguards - Anthropic (August 7, 2026)