Anthropic broadens cyber defense access to top AI models

Artificial intelligence models designed to block cyberattacks can sometimes hinder legitimate security researchers who need to test corporate defenses. To address this issue, Anthropic announced an update to its Cyber Verification Program, creating three distinct access tiers that grant vetted security professionals broader permission to use advanced models like Claude Opus 5.5 for defensive research.

Because software tools used to repair vulnerabilities can often be repurposed by hackers, leading AI developers normally place strict safeguards on code related to digital security. By relaxing these automatic blocks for approved organizations, Anthropic aims to help security operations centers, critical infrastructure operators, and authorized penetration testing teams detect software flaws before cybercriminals can exploit them.

However, granting specialized access introduces new operational risks if defensive tools fall into the wrong hands or if automated screening fails to detect malicious intent. According to Anthropic, participating organizations must allow temporary data retention for safety monitoring until self-hosted privacy options launch later in the year. It remains unclear whether this tiered verification framework will become an industry standard for managing dual use AI technologies, or if maintaining strict control over custom access levels will prove too burdensome for providers to scale safely across thousands of security firms.