Anthropic opens access to advanced AI models for cybersecurity teams

Anthropic has expanded its Cyber Verification Program, allowing cybersecurity professionals to test its most powerful AI models with reduced security restrictions. The initiative, which includes the previously launched Glasswing Project, has already identified over 129,000 verified software vulnerabilities since April. Anthropic estimates the actual number of vulnerabilities could be five times higher due to limited data collection. The new program, launched on October 7, integrates two existing projects and offers three tiers of access based on organizational qualifications. The highest tier allows testing of critical systems like power grids and aviation infrastructure. Anthropic, in collaboration with the U.S. government, will review all participants' credentials, with existing Glasswing members automatically placed in the highest tier.