Anthropic recently released Claude Fable 5, touting it as the most powerful AI model it has made widely available, with particular praise for its biology capabilities. However, the model refuses to answer basic biology questions that a high school student could handle, instead deferring to the previous flagship model, Claude Opus 4.8.
Why Fable Won't Answer
This limitation is not due to a lack of knowledge but is by design. Fable is a public-facing Mythos-class model, a family so capable in cybersecurity that Anthropic initially deemed it too dangerous for public release. While the company has focused on cybersecurity risks, biology is where the guardrails are most obvious and restrictive.
During testing, Fable refused to answer questions about cell membranes, mitochondria, prions, and how mRNA vaccines work. It also declined to explain hay fever causes, asthma medicine, antibiotic resistance, or Ebola. Some basic queries like "what is cancer" and "what is DNA" occasionally got through.
Conservative Safeguards
Anthropic spokesperson Paruul Maheshwary stated that the broad biology filters are intentional and deliberately conservative, primarily to prevent bioweapons-related misuse. The company believes that with Mythos-class models, the potential for real-world scientific tasks also increases the risk for malicious actors.
Anthropic has identified four key areas for throttling responses: chemistry, biology, cybersecurity, and distillation. While Fable was more willing to answer chemistry and cybersecurity questions, it still deferred on highly toxic agents like sarin gas.
Future Plans
Maheshwary acknowledged that the mitochondria refusal appears to be a false positive and that Anthropic is working to improve detection. The company intends to make Mythos-class models available without these safeguards to the broader biology and life sciences community to accelerate biomedical research.
Anthropic did not clarify whether this restricted release will become the norm for future models.
