Anthropic cut biology false positives by 85% in Fable 5—but virology stays locked
Curated by the Inblix editorial team
Anthropic has dramatically loosened Fable 5’s biology restrictions, slashing false positives in its safety filters by roughly 85 percent. The change is a direct response to months of sharp criticism from the scientific community. Previously, Fable 5’s overzealous filters blocked nearly all biology queries and silently rerouted them to the less capable Opus 5, frustrating researchers who just wanted to interpret lab results or look up symptoms. Now, those routine medical and biological questions can be handled by the frontier model.
But the lab door isn’t wide open. Anthropic is keeping strict guardrails on what it calls “dual-use” topics—virology, toxicology, and molecular design. The company argues that Fable 5 could give bad actors capabilities they can’t get anywhere else, and a released pathogen doesn’t come with an off switch. That stance is backed by more than just corporate anxiety. An analysis by U.S. intelligence agencies, a review of ChatGPT chats seeking bioweapon instructions, joint warnings from researchers, and a Stanford project that used AI to design fully synthetic viruses all point to a threat that’s real and hard to contain.
For researchers who need access to those restricted areas, Anthropic is building access programs so they can eventually use the locked features. It’s a familiar tension in AI safety: how do you give scientists the tools they need without handing a loaded weapon to someone who shouldn’t have it? The difference here is that biological risks don’t behave like cyberattacks. You can patch a server or revoke an API key. Countermeasures for a virus take time—and that time costs lives.
The 85 percent reduction in false positives is a meaningful concession to the research community, and it suggests Anthropic’s safety systems can be tuned with more precision than critics assumed. But the core debate hasn’t shifted. The company is betting that virology and toxicology are qualitatively different from general biology in a way that justifies permanent restrictions. Whether that line holds—or whether it’s just the next battlefront for scientists who feel locked out of their own field—is the question that’ll define the next round of this fight.
💡 Key Takeaways
- Anthropic reduced Fable 5's biology false positives by 85%, letting the frontier model handle routine medical queries that were previously blocked and routed to the weaker Opus 5.
- Restrictions stay locked on virology, toxicology, and molecular design—topics where Anthropic argues bad actors could gain genuinely novel capabilities.
- U.S. intelligence analysis and a Stanford project that used AI to design synthetic viruses give weight to the risk, distinguishing biological threats from contained cybersecurity incidents.
Keep reading: See related articles below for more coverage on this topic.
Get smarter about AI
The sharpest AI news, curated daily. Delivered free to your inbox.