Anthropic’s Claude Fable 5 draws fire for blocking basic biology questions

Anthropic launched Claude Fable 5 on Tuesday, its first publicly available Mythos-class model, but safety classifiers silently reroute flagged queries to the older Opus 4.8.techcrunch
The Verge reported the model refused to explain hay fever, antibiotic resistance, and asthma medication, while a cancer researcher on Reddit said every query was blocked.theverge
Anthropic called the broad filters a deliberate trade-off to prevent bioweapons misuse and said it is working to reduce false positives.theverge

Claude Fable 5 Safety Guardrails Draw Criticism for Blocking Mundane Queries

Anthropic launched Claude Fable 5 on Tuesday, its first publicly available Mythos-class AI model, but the system’s automated safety classifiers are already frustrating users who say the guardrails block routine questions about biology, medicine, and cybersecurity — topics the model was specifically marketed as excelling in.

Broad Filters, Narrow Access

Fable 5 shares its underlying architecture with Claude Mythos 5, a restricted model available only to vetted partners. The key difference is a layer of classifiers that monitor requests involving cybersecurity, biology, chemistry, and model distillation. When triggered, responses are silently routed to Claude Opus 4.8, Anthropic’s previous top model. Anthropic says the fallback fires in fewer than 5 percent of sessions on average.techcrunch

But that average obscures a sharper problem. The Verge reported Wednesday that Fable refused to answer basic biology questions a high school student could handle, including queries about hay fever, antibiotic resistance, and how asthma medication works. On Reddit, a cancer researcher said that because their account profile mentions terms like “prostate cancer cell lines,” every single query — including greetings and weather questions — gets rerouted to Opus 4.8. Anthropic spokesperson Paruul Maheshwary told The Verge the company “made this trade so customers benefit from the model’s capabilities sooner without the risks”.reddit

Industry Reaction

Business Insider noted the restrictions have drawn criticism from analysts and early users. Patrick Moorhead, CEO of Moor Insights & Strategy, wrote on X that Fable 5 refused to help with earnings-analysis talking points and a board presentation because it deemed the tasks “too dangerous”. “Is this the model we’re all frightened of and makes Anthropic worth $1T?” he wrote.businessinsider

Meanwhile, Microsoft is reportedly restricting employee use of Fable 5 over concerns about Anthropic’s mandatory 30-day data retention policy for all Mythos-class model traffic, according to The Verge.usnews

A Deliberate Trade-Off

Anthropic has been transparent that the biology filters are intentionally conservative. The company said Fable 5 is the first Mythos-class model it believes has “the ability for bad actors to use for highly risky biological research,” and that an overly cautious approach was necessary for safe deployment. The company says it is working to reduce false positives and plans to extend unrestricted Mythos 5 access to broader life sciences researchers through a trusted access program.anthropic

Through June 22, Fable 5 remains included at no extra cost for Pro, Max, Team, and Enterprise subscribers. After that, access will require usage credits.thenextweb