Anthropic expands AI model access for government and cyber defense partners

Investing.com -- Anthropic (NASDAQ:ANTP) has expanded access to its most sophisticated artificial intelligence models, consolidating its security initiatives to provide vetted cyber defense teams and select government partners with advanced tools to safeguard critical infrastructure.

The San Francisco-based AI startup unveiled an expanded Cyber Verification Program on Tuesday, integrating its Project Glasswing initiative into a streamlined, three-tier framework. Under the updated structure, qualifying security professionals will gain calibrated access to Anthropic's most capable models, including Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1.

The move addresses the inherently dual-use nature of cyber capabilities, balancing conservative safeguards on public models with targeted access for high-risk security work. By lowering automated blocking classifiers for verified users, Anthropic aims to bolster defensive posture against sophisticated threats targeting power grids, financial networks, and transportation systems.

The program establishes three distinct access levels, Defense, Red Team, and Specialized, tailored to specific organizational roles and security control requirements. The Specialized tier, managed in close collaboration with the U.S. government, grants the broadest access for stress-testing systems where disruption could carry severe real-world or market consequences.

Internal benchmark testing on the CyScenarioBench evaluation demonstrated that the updated safeguards effectively eliminate operational friction for authorized red-teaming while maintaining strict controls at lower tiers. In testing with Claude Opus 5.5, the Red Team Access tier achieved a 68% task completion rate with zero automated blocks, matching the performance of unconstrained models.

The expansion builds on significant operational momentum from Project Glasswing, which helped partner organizations identify more than 129,000 verified software vulnerabilities between April and July. Participating entities, including Booz Allen and Comcast, reported that the advanced AI capabilities accelerated vulnerability detection timetables by months or years.

To balance expanded functionality with enterprise security risks, Anthropic is requiring mandatory data retention for program participants to monitor for potential misuse. Organizations will gain the ability to store data in their own cloud infrastructure later this fall following the launch of Enterprise Frontier Safeguards, with broader deployment planned across major cloud platforms including Google Cloud Vertex AI and Microsoft Foundry.

Anthropic expands AI model access for government and cyber defense partners

Black Hills stock surges 5% on $1.8B Google data center deal

Sigma Lithium stock rises after court upholds mining licenses