Myth will not answer basic biology questions

Anthropic only Claude Fable 5 was releasedHe described it as the most powerful AI model ever made widely available to praise Her skills are in biology, among other things. But the model won’t answer basic biology questions, the kind you’d expect a high school student to grapple with. Instead, it delivers the query to Previous main modelClosing of business 4.8

Not that Fable doesn’t know the answers. That’s because Anthropic, by design, won’t allow it.

Fable is a public-facing model of the Mythos class, a family very capable of cybersecurity missions. Anthropic said it was Too dangerous to be published publicly. But while Anthropic spent a lot of… Extended Mythos rollout As a warning about cybersecurity, biology is where Fable’s guardrails are most evident — and most restrictive.

When I tried the model, it refused to answer a host of basic biological questions, many of which seemed as far removed from any potential safety risk as any. He will not respond to “Tell me about cell membranes” or answer “What are mitochondria,” that famous powerhouse of the cell. She refused to explain “what a prion is,” or the protein particles that cause mad cow disease, or “how mRNA vaccines work.”

The restrictions apply to mundane and fairly objectively innocuous medical inquiries as well. The story won’t answer what causes hay fever, explain how asthma medications work, explain how antibiotic resistance arises, or tell me what Ebola is and how it spreads. Sometimes I’m able to answer some of my basic queries, as Fable answers questions like “what is cancer” and “what is DNA”. When I turned down Fable, the Opus 4.8 responded quite well.

Anthropic says the broad biology filters are an intentional choice and are intentionally conservative, with biological weapons being the primary concern. “With the launch of Claude Fable 5, our first Mythos-class model, we believe models now have a greater ability to accomplish real-world scientific missions and malicious actors can use our models in high-risk biological research,” spokesman Parul Maheshwari said. Edge. “We have always used classifiers to prevent our models from assisting with requests related to biological weapons. To safely deploy Fable 5, we believe it was necessary to be overly conservative with our safeguards, such that they block most queries related to biological work.”

Anthropic has already mentioned Highlight Four main areas would stifle Fable’s safety responses: chemistry, biology, cybersecurity, and distillation, a technique for training smaller AIs using the output of larger AIs. The company has accused Chinese competitors like DeepSeek use distillation in their models on an “industrial” scale.

Although I wasn’t able to meaningfully test distillation, Fable seemed more than willing to answer questions about chemistry and cybersecurity. For example, it provided a basic overview of the explosive TNT, although assembly instructions were withheld “for obvious reasons.” He easily answered questions about the use of chlorine gas as a chemical weapon, common password threats, and nuclear fusion and fission, as well as explaining how to secure your iPhone from hackers. Still Limited : The anecdote is deferred to Opus when she asks her about sarin, a highly toxic nerve gas. Both Fable and Opus dismissed the ‘how to make anthrax’ question, and Claude paused the chat altogether. This makes sense. Immediate mitochondrial rejection appears to be a false positive result.

“We made this trade-off so that customers can benefit from the model’s capabilities sooner without risk,” Maheshwari explained, adding that Anthropic is working hard to improve its detection and reduce false positives. “We intend to make Mythos class models available without these safeguards to the broader biology and life sciences community so that these capabilities can be used to accelerate biomedical research and drug discovery.”

Anthropic did not answer questions about whether this type of restricted release will become the new standard for future models.