TLDR
- Anthropic has launched a three-tier framework granting qualified cybersecurity firms access to its cutting-edge Claude AI models.
- The Cyber Verification Program features Defense Access, Red Team Access, and Specialized Access levels with varying permissions.
- JPMorgan Chase’s Jamie Dimon stated the Mythos model elevated worldwide cybersecurity threats by a factor of ten.
- Security partners leveraging Claude Mythos identified over 129,000 confirmed software flaws from April through July 2026.
- The company will introduce Enterprise Frontier Safeguards this autumn to balance data privacy with abuse monitoring.
Anthropic has broadened the scope of its initiative to share sophisticated artificial intelligence capabilities with qualified cybersecurity professionals. The firm revealed this development on Tuesday.
This enhanced initiative merges two previous pilotsāProject Glasswing and the original Cyber Verification Programāinto a unified framework featuring three distinct access levels.
The AI systems available through the program include Claude Opus 5.5, Claude Sonnet 5.5, and Claude Mythos 5.1. Additional models will become available upon their release.
The company initially provided restricted access to its Mythos system in April. Recipients included government bodies, banking institutions, and leading software enterprises.
How the New Access Tiers Work
Defense Access represents the entry level. This tier targets security professionals engaged in protective activities, including incident response and vulnerability remediation.
Red Team Access, the second tier, permits sanctioned penetration testing. Teams at this level can evaluate systems they’re authorized to assess, including critical infrastructure assets.
Specialized Access, the most permissive tier, carries the least restrictions. Organizations collaborating with U.S. government entities to examine systems like electrical grids, aviation operations, and financial transfer networks qualify for this level.
The company emphasized that safeguards preventing physical harm or widespread disruption remain active across all tiers, including the highest access category.
Current Project Glasswing participants will automatically transition to Specialized Access status. They won’t need to submit new applications for existing models.
Rising Concerns Over Cyber Risk
This broadened access follows mounting warnings from leaders about AI-related cybersecurity challenges. JPMorgan Chase CEO Jamie Dimon characterized Mythos as having amplified global cyber risks by a factor of ten.
In August, the U.S. government announced partnerships with private sector companies to address overseas cyberattacks. This marked an expansion of activities previously dominated by government agencies.
The company conducted testing to evaluate the effectiveness of protections at each tier. Without program enrollment, every test attempt encountered immediate blocking.
Within the Defense Access tier, 46 out of 50 test scenarios faced blocking at some stage. The Red Team Access tier saw zero blocked trials.
According to Anthropic, organizations using Claude Mythos discovered more than 129,000 confirmed software vulnerabilities during a four-month span from April to July 2026. Over 33,000 received critical or high severity classifications.
Internal scanning operations conducted by the company uncovered an additional 5,500 vulnerabilities from April through October 2026. Anthropic acknowledged the actual total likely exceeds these figures.
This fall, the company intends to deploy Enterprise Frontier Safeguards. The system aims to enable organizations to maintain private data storage while permitting Anthropic to detect misuse.
Organizations holding zero data retention agreements can continue program participation under existing privacy conditions until the new system launches. The company committed to ongoing refinement of its protective measures.





