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

Anthropic’s internal filtering system for biological and chemical weapons risks was inactive for nearly a year, exposing 133 million unfiltered interactions with its AI models. This revelation comes from a safety report released by Anthropic, which also involved feedback from 50,000 external contractors. The downtime raises significant concerns regarding the safety protocols in place for AI interactions, particularly in sensitive areas like bio-weapons. This follows previous reports on Anthropic’s safety measures and their implications for AI deployment in critical sectors. Practitioners should reassess reliance on Anthropic’s models for applications involving safety-sensitive content, given this lapse in filtering capabilities. The Decoder reported.