Government surveillance agencies attempted to leverage Claude's AI capabilities to optimize their global operations, Anthropic said in a report released Thursday. The company identified attempts to misuse the model for cyberattacks, influence operations, biology, and building weapons, though it disrupted every operation featured in the findings. These cases were used to strengthen the model's security safeguards.

The threat intelligence report was shared with authorities and other AI firms to help identify similar patterns across different platforms. Anthropic characterized the reported incidents as sophisticated outliers that demonstrate the trajectory of emerging threats rather than typical user behavior.