Anthropic gives more security teams access to Claude with fewer safety restrictions
decoder ·
Anthropic has expanded its Cyber Verification Program to provide vetted security professionals with access to Claude's most powerful AI models, featuring reduced safety filters. This initiative supports vulnerability research, malware analysis, incident response, and penetration testing, which are typically blocked in public versions of the model due to potential misuse risks. The program is structured into three distinct tiers: Defense Access for broad organizational use, Red Team Access for authorized attack simulations, and Specialized Access for critical infrastructure testing involving government vetting. Previous iterations of this effort, specifically Project Glasswing, yielded over 129,000 confirmed vulnerabilities between April and July 2026. More than 33,000 of these findings were classified as high-severity or critical issues. Anthropic estimates that the real-world impact of these discoveries is at least five times higher than the reported numbers. Additionally, several partners noted that the AI significantly accelerated their workflow by months or even years.
decoder