WorkThe story, in brief
Anthropic's bio-weapons filter was down for nearly a year, exposing 133 million requests
133 million unfiltered requests. Anthropic's bio-weapons safety filter was down for nearly a year—and it just disclosed it in a safety report.
The KeyNews take
Why it matters
A major frontier lab's critical safety system failed silently for months, exposing the gap between public safety commitments and operational reality. This is both a governance failure story and a cautionary tale for practitioners deploying AI at scale.
The key facts
5 to knowAnthropic's bio-weapons/chemical weapons filter inactive for ~12 months
133 million unfiltered model interactions during outage
50,000 external feedback contractors affected
Disclosure came via internal safety report
Safety system failure at a frontier lab with public safety positioning
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: In a safety report, Anthropic reveals that its internal filtering system for biological and chemical weapons risks was inactive for nearly a year. During that time, around 50,000 external feedback contractors ran about 133 million unfiltered interactions with the models.