Anthropic Updates Claude Fable 5 Bio-Safety, Cutting False Blocks by 85%

Anthropic, a US AI startup, recently announced updates to the bio-safety mechanisms of its Claude Fable 5 model, greatly reducing false blocks. Users asking biology questions now see fewer downgrades, with official tests showing an 85% drop in such cases, allowing the model to handle a wider range of biology tasks.

To prevent misuse for activities like bioweapon research, Anthropic had set up safety classifiers for Claude to detect requests for protected biology tasks or harmful content. The update adjusts these classifiers to better tell harmless questions from high-risk ones.

Still, to keep AI risks in check, Fable 5 continues to limit requests related to professional biology research and drug development.