Real-time cyber safeguards on Claude Opus and Sonnet (support.claude.com)

Anthropic has announced the implementation of new real-time cyber safeguards for its Claude Opus and Sonnet models to enhance security and compliance with its Usage Policy. These safeguards automatically detect and block potentially high-risk cybersecurity requests, categorizing activities into prohibited uses—like data exfiltration and ransomware development—and high-risk dual use, which may have legitimate applications in cybersecurity. While the latter is blocked by default, professionals can apply through the Cyber Verification Program (CVP) to access these tools safely for legitimate defensive purposes. The significance of this update lies in its potential to create a safer environment for cybersecurity practitioners while maintaining ethical boundaries in AI usage. By evaluating emerging threats continuously, Anthropic aims to evolve these safeguards alongside technological advancements. The CVP allows legitimate users to navigate the restrictions, thereby minimizing disruption to crucial cybersecurity work while ensuring that malicious intents are effectively curtailed. As this initiative progresses, Anthropic pledges ongoing refinement of the safeguards, reinforcing its commitment to responsible AI practices.