The Information Machine
Following·New·first covered 6 Oct 2026·4 sources

Anthropic Expands Cyber Verification Program With Three Access Tiers

The gist

The program gives vetted security professionals access to reduced-safeguard AI capabilities for offensive and defensive work that would otherwise be blocked. The Project Glasswing results give a concrete data point for the volume of vulnerabilities AI-assisted security work can surface.

The full picture

Anthropic launched an expanded Cyber Verification Program on October 6, 2026, consolidating Project Glasswing and its prior verification program into three access tiers: Defense Access, Red Team Access, and Specialized Access. Verified participants gain access to Claude Mythos 5.1, Opus 5.5, and Sonnet 5.5. Individuals may apply only for Defense Access and must be on a paid plan; Red Team and Specialized Access are open to organizations only. Each organization files a single application, and Anthropic places the applicant at the highest tier the provided information supports, with responses targeted within seven business days. Eligibility factors include the nature of the organization's work, Anthropic's ability to verify identity, the legal and regulatory environment, and risk of diversion or compelled access. Anthropic said it takes a more cautious approach for organizations primarily serving military, intelligence, or law enforcement customers.

The program unlocks high-risk dual-use activities such as vulnerability exploitation and offensive tooling for verified participants, but does not unlock outright prohibited uses. Ransomware development and mass data exfiltration remain blocked for all organizations regardless of verification status.

Through Project Glasswing, partner organizations uncovered at least 129,000 verified software vulnerabilities between April and July 2026, with more than 33,000 rated critical- or high-severity. Several partners said the Claude Mythos models had increased their rate of vulnerability finding by months or even years. Benchmark testing on CyScenarioBench showed that without CVP access every task was blocked, while in the Red Team Access tier no blocks occurred and Claude Opus 5.5 completed 34 of 50 tasks, matching its baseline 67.6% success rate.

Data retention is required for program participants to allow misuse monitoring. An upcoming Enterprise Frontier Safeguards solution will eventually allow eligible organizations to store data in their own cloud infrastructure, though organizations on Zero Data Retention agreements are not currently eligible. The program is available on the Claude Platform, Google Cloud Vertex AI, and Microsoft Foundry; Amazon Bedrock access is limited to Enterprise Frontier Safeguards-eligible customers.

How it developed
6 October 2026

Anthropic launches expanded Cyber Verification Program, consolidating Project Glasswing and prior CVP into three tiers: Defense Access, Red Team Access, and Specialized Access

Sources
The daily email

Want this in your inbox?

I send one email each morning with the stories that moved. If you would rather just read here, that works too.

Subscribe free