Anthropic's bio-weapons filter was down for nearly a year, exposing 133 million requests

Anthropic disclosed that its internal safety filters designed to prevent the generation of biological and chemical weapon information were inactive for nearly a year. This configuration error allowed external contractors to conduct 133 million model interactions without those specific security guardrails in place. While the company stated no safety incidents occurred, the lapse highlights the operational challenges of maintaining robust oversight systems during the continuous development and deployment of generative artificial intelligence.