Anthropic Grants 'Red Team Access' to Claude: Security Organizations Authorized for Penetration Testing

Beating AI Newsflash: Anthropic has expanded its Cyber Verification Program (CVP), merging the previous single-tier access and Project Glasswing into a three-tier system. Approved security teams can now access Claude Opus 5.5, Sonnet 5.5, and Mythos 5.1 with reduced cybersecurity restrictions. The newly added Red Team Access explicitly allows authorized penetration testing and red team exercises. The three tiers are Defense, Red Team, and Specialized. Defense is designed for incident response, malware reverse engineering, and vulnerability analysis, and individual security researchers can also apply. Red Team is exclusively available to organizations, enabling them to test authorized systems. Specialized has the least restrictions and is primarily intended for rigorously vetted institutions to test critical systems such as power grids, aviation systems, telecommunications networks, and banking transfers. Anthropic ran 50 complex cyberattack tasks using Opus 5.5. In the standard version, all 50 attempts were intercepted in the first round; under the Defense tier, 46 attempts were still intercepted mid-process. However, under the Red Team tier, none of the 50 attempts triggered security interceptions, with 34 tasks successfully completed—comparable to the success rate when such security restrictions are entirely disabled. Nevertheless, operations that could cause severe harm, such as ransomware or physical system destruction, remain strictly prohibited. Previously, Anthropic only offered Mythos' advanced cybersecurity capabilities to select institutions through Project Glasswing. Between April and July, partners identified at least 129,000 software vulnerabilities, over 33,000 of which were classified as high-risk or critical. Now, Glasswing has been integrated into CVP, and existing members are automatically upgraded to the highest Specialized tier. [Original Link]