Anthropic Reduces False Positives in Fable 5

Anthropic Reduces False Positives in Fable 5

Anthropic has reduced false positives in its biological safety filters for Fable 5 by approximately 85%. Previously, nearly all biological queries were blocked and redirected to the less capable Opus 5, which drew sharp criticism from scientists. Users can now handle many more biological tasks with Fable 5, such as interpreting lab results, understanding symptoms, and answering medical questions.

The classifier now allows most benign biological queries to pass while blocking dual-use research topics.

Restrictions remain for dual-use subjects such as virology, toxicology, and molecular design, where Anthropic claims that Fable 5 could provide malicious actors with capabilities unavailable elsewhere. The company is also developing access programs so that researchers can eventually use these restricted features.

Anthropic's justification: biological risks can be more difficult to contain than cyberattacks. A released virus cannot be stopped, and countermeasures take time. An analysis from U.S. intelligence agencies, a review of ChatGPT discussions requesting instructions on biological weapons, joint warnings from researchers, and a Stanford project that used AI to design entirely synthetic viruses all suggest that the threat is real and not merely a reflection of Anthropic's usual anxiety.