A recently disclosed safety report from AI developer Anthropic reveals that its internal filtering system for detecting biological and chemical weapons risks was not functioning for nearly a year. During this period, external contractors processed approximately 133 million interactions with the company's models without the filter in place. The report highlights the potential risks associated with this prolonged outage, which may have compromised the security of the AI system. This incident underscores the importance of robust safety protocols in AI development to prevent unintended consequences.