Anthropic's Claude Fable 5 Refuses Basic Biology Questions Over Bioweapon Concerns
Key Takeaways
- ▸Claude Fable 5 refuses to answer even basic, harmless biology questions like 'what are mitochondria' to prevent potential bioweapon misuse
- ▸Anthropic acknowledged the safeguards are 'overly conservative,' blocking most biology-related queries and deferring them to the older Claude Opus 4.8 model
- ▸The restrictions extend to four key areas: chemistry, biology, cybersecurity, and distillation—with biology showing the most severe limitations
Summary
Anthropic has deliberately implemented overly conservative safeguards in Claude Fable 5, its first publicly released Mythos-class model, that prevent the AI from answering basic biology questions—even those far removed from any plausible safety risk. Queries about cell membranes, mitochondria, mRNA vaccines, antibiotic resistance, and hay fever are all rejected, with the model deferring to the older Claude Opus 4.8 instead. The company justified the restrictions as necessary to prevent malicious actors from using the model for bioweapon development, stating it needed to "be overly conservative with our safeguards so they block most queries tied to biology work." Anthropic has also implemented similar restrictions in chemistry, cybersecurity, and distillation research, though the biology filters appear most restrictive. The tradeoff raises questions about balancing AI safety with accessibility to legitimate scientific and educational information.
- The company prioritized faster public deployment of Fable 5's capabilities over more permissive safety standards
