Anthropic is updating Claude Fable 5's biology-related safety filters to reject fewer legitimate queries, while warning that the same capabilities could lower the barrier for biological weapons development.
The company said users will experience roughly 85% fewer biology-related "fallbacks", instances where the model refuses to answer a question it could safely handle. The change reflects a calibration tradeoff that every frontier lab faces: overly aggressive filters block legitimate research and education, while permissive ones risk enabling misuse.
The Tradeoff
Anthropic's announcement notes that Fable 5 is more capable of handling complex biological tasks than its predecessors, and that those capabilities, if misused, could aid malicious actors in developing biological weapons. The company said "more work is needed to refine our safeguards" and that it is monitoring for signs of abuse.
Industry Context
The update is part of a broader pattern across the industry. Labs have been tightening and loosening safety filters in different domains as they accumulate real-world usage data. OpenAI's GPT-5.6 Sol, for example, was recently flagged by its own safety team for potentially reaching critical cybersecurity risk levels after a separate evaluation.
The Fable 5 biology update rolls out to users immediately. Anthropic did not disclose whether the changes apply to the API, the consumer chat interface, or both.
Sources
Times Now: Anthropic says Claude Fable 5 will stop rejecting biology questions, warns of bioweapon risks - https://www.timesnownews.com/technology-science/anthropic-says-claude-fable-5-will-stop-rejecting-biology-questions-warns-of-bioweapon-risks-article-155361466/amp
Inshorts: Anthropic says Claude Fable 5 will reject fewer biology queries, warns of bioweapon risks - https://inshorts.com/prev/en/news/anthropic-says-claude-fable-5-will-reject-fewer-biology-queries--warns-of-bioweapon-risks-1786177400344
