Anthropic expands Project Glasswing to more security teams

On October 6, 2026, Anthropic expanded its Cyber Verification Program, merging its Project Glasswing initiative and original verification program. The expansion grants vetted security teams access to advanced models like Claude Opus 5.5 and Claude Mythos 5.1 with reduced safeguards. In collaboration with the US government, the program's specialized tier allows high-risk offensive testing of critical infrastructure, including power grids and banking systems, following the discovery of over 129,000 vulnerabilities.

The expanded Cyber Verification Program

• ▪Anthropic's expanded Cyber Verification Program features three distinct tiers, each with its own verification requirements and security controls, to manage access to Anthropic's advanced artificial intelligence models
• ▪On October 6, 2026, Anthropic announced the expansion of its Cyber Verification Program, combining its Project Glasswing initiative and the original Cyber Verification Program to allow vetted cybersecurity professionals to test its most powerful AI models with fewer safeguards
• ▪All three tiers—Defense, Red Team, and Specialized—of Anthropic's expanded Cyber Verification Program include access to Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and future artificial intelligence models

The Defense tier of the program

• ▪Security teams, critical infrastructure operators, open-source maintainers, and researchers with a documented record of reported vulnerabilities are eligible to apply for the Defense tier of Anthropic's expanded Cyber Verification Program
• ▪The Defense tier of Anthropic's expanded Cyber Verification Program covers cybersecurity work such as incident response and malware analysis

The Specialized tier of the program

• ▪The Specialized tier of Anthropic's expanded Cyber Verification Program has the fewest restrictions, reserving high-risk offensive testing on safety-critical systems—including power grids, flight operating systems, and interbank transfer infrastructure—for a small group of organizations
• ▪Anthropic vets each member of the Specialized tier of its expanded Cyber Verification Program in collaboration with the United States government

Vulnerabilities discovered by Project Glasswing

• ▪More than 33,000 of the software vulnerabilities discovered through Anthropic's Project Glasswing initiative in 2026 have been rated as critical or high severity
• ▪Anthropic's Project Glasswing partners uncovered at least 129,000 verified software vulnerabilities between April and July 2026, while Anthropic's open-source scanning found an additional 5,500 vulnerabilities between April and October 2026

Claude Mythos models

• ▪Anthropic's Project Glasswing initiative previously provided organizations securing critical software with access to Claude Mythos, Anthropic's most cyber-capable family of models
• ▪Anthropic's April 2026 unveiling of Claude Mythos Preview raised public fears that artificial intelligence could be used to hack software before it had been secured

Debatable claims

• ▪In the current AI era, offensive cyber capabilities have a near-term advantage over defensive measures
• ▪High-risk offensive AI testing on critical infrastructure does more harm than good
• ▪Government vetting of private AI researchers is an appropriate way to protect national security
• ▪Anthropic should not lower safety safeguards for external cybersecurity testers