Anthropic Expands Claude Access for Cyber Defense Teams

Anthropic has expanded access to its most advanced artificial intelligence models for verified cybersecurity professionals and select government partners, aiming to strengthen defenses against threats targeting critical infrastructure.

The San Francisco-based AI company announced an expanded Cyber Verification Program on Tuesday, folding its previous Project Glasswing initiative into a three-tier access framework. Qualified security teams can receive calibrated access to Anthropic’s most capable AI models, including Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1.

Anthropic said the program is designed to address the dual-use risks associated with powerful AI cybersecurity capabilities. While public models will continue operating under conservative safeguards, verified professionals can receive fewer automated restrictions when conducting legitimate defensive security work.

The framework includes Defense, Red Team and Specialized access levels, with permissions tailored to organizational responsibilities and security requirements. The Specialized tier offers the broadest capabilities and is administered in close coordination with the U.S. government. It is intended for testing systems where failures or disruptions could have significant real-world or financial consequences.

Anthropic’s CyScenarioBench testing indicated that the revised safeguards can reduce obstacles for authorized cybersecurity teams without broadly weakening protections. Claude Opus 5.5 achieved a 68% task completion rate under Red Team Access with no automated blocks, matching the performance recorded by unrestricted models.

The expansion follows results from Project Glasswing, which helped participating organizations uncover more than 129,000 verified software vulnerabilities between April and July. Partners including Booz Allen and Comcast said Anthropic’s advanced AI tools significantly accelerated vulnerability discovery, potentially reducing timelines by months or even years.

Anthropic is also introducing additional enterprise safeguards as access expands. Participants must comply with mandatory data-retention requirements intended to help detect potential misuse.

Later this fall, organizations are expected to gain options for storing program data within their own cloud infrastructure through Enterprise Frontier Safeguards. Anthropic also plans broader availability through major cloud platforms, including Google Cloud Vertex AI and Microsoft Foundry.