Anthropic bio-weapons filter outage exposed 133 million contractor requests

Anthropic revealed in a safety report that its internal filtering system for biological and chemical weapons risks was inactive for nearly a year. The report says the blocking biological classifiers were inactive from May 2025 through April 2026. These filters are designed to prevent AI models from being used to extract dangerous knowledge about chemical or biological weapons.

The outage affected external contractors providing human feedback, not just a small internal test group. According to the article, about 50,000 people ran roughly 133 million chats with the models while the filters were inactive. Anthropic said these individuals were vetted only by external vendors, whose screening processes were often insufficient.

Anthropic says its internal investigation found no evidence of actual misuse during the period when the classifiers were inactive. The company has since tightened contractor requirements. The key takeaway for readers: even safety-focused AI systems depend on operational controls staying active and verified over time.

The lapse is notable because it concerns safeguards for biological and chemical weapons risks, an area Anthropic has publicly treated as serious. The same report also lands amid a broader tension: filters must block dangerous requests without preventing legitimate research. The article notes that Anthropic recently loosened classifiers on Fable 5 after researchers complained that earlier filters were too aggressive.