Anthropic's safety filter designed to block bioweapons-related queries on its Claude AI assistant was disabled for nearly a year, potentially exposing the system to requests seeking information on creating biological weapons, according to a report by The Decoder.
What Happened
The company revealed that its bioweapons content filter had been down since August 2025 and was only restored in July 2026. During this period, approximately 133 million user requests passed through the system without the safety screening. Anthropic has not disclosed specific details about what types of queries were made during the outage or whether any harmful content was generated as a result.
Why It Matters
The extended failure of Anthropic's bioweapons filter raises questions about the company's safety monitoring and incident detection processes. Biological weapons development information represents one of the most severe risks in AI safety, and the ability to screen such requests is considered a critical safeguard by many researchers and policymakers. The outage also comes as governments worldwide are developing regulations requiring AI companies to implement robust safety measures for frontier models.
The Bottom Line
Anthropic has restored the bioweapons safety filter and states it has implemented additional monitoring systems to detect similar failures more quickly in the future. The company reports that during the outage, standard abuse detection mechanisms remained active, though these are designed to catch different types of misuse than specialized content filters.