Anthropic's bio-weapons filter was down for nearly a year, exposing 133 million requests
Anthropic disclosed its biological and chemical weapons safety filter remained inactive for nearly a year. During this period, approximately 133 million unfiltered requests were processed by external contractors.
Anthropic published a safety report detailing a lapse in their internal filtering mechanisms designed to prevent misuse of their models for biological or chemical weapons development.
The report indicates the filter was disabled for close to twelve months. Consequently, around 50,000 contractors engaged in roughly 133 million interactions with the models without this specific safety layer active.
This incident highlights ongoing challenges in AI safety governance and the complexity of maintaining robust guardrails during model development and evaluation phases.
Such disclosures underscore the risks associated with dual-use technologies and the necessity for continuous monitoring systems to mitigate potential harm from advanced language models.
This page provides an editorial summary based on publicly available information. It is not a republished article. Use the source link below for the original report.