Anthropic Expands Vetted Cyber Access as Glasswing Reports 129,000 Vulnerabilities
Summary
Anthropic is expanding its Cyber Verification Program for vetted cybersecurity teams. The company says Project Glasswing verified at least 129,000 vulnerabilities from April to July 2026, while separate research found only two of 300 reviewed flaws exploited in the wild
Key points
- Anthropic’s Cyber Verification Program offers three access tiers for vetted organizations and security teams to test its advanced AI models with different safeguard levels.
- Anthropic says Project Glasswing found at least 129,000 verified software vulnerabilities between April and July 2026, including more than 33,000 rated critical or high; it warns the count may be an undercount.
- The company reports finding an additional 5,500 verified vulnerabilities through open-source scanning between April and October 2026.
- In a CyScenarioBench evaluation, Claude Opus 5.5 blocked 46 of 50 tasks in the Defense Access tier; Red Team Access blocked none and completed 34.
- VulnCheck researcher Patrick Garrity reported that two of 300 vulnerabilities attributed to Anthropic or Project Glasswing had been exploited in the wild: CVE-2026-26980 and CVE-2026-61500.
- The article cautions that AI-discovered flaws are not necessarily exploitable and that AI-generated code and patches can introduce security risks.
Article Details
- Event Type
- Anthropic expanded its Cyber Verification Program, giving vetted cybersecurity teams tiered access to AI models with reduced safeguards. Anthropic also reported vulnerability findings from Project Glasswing and open-source scanning.
- Impact
- Anthropic said Project Glasswing found at least 129,000 verified software vulnerabilities between April and July 2026, with another 5,500 found through open-source scanning between April and October. More than 33,000 were rated critical or high; Anthropic said this was likely an undercount and estimated the true impact could be at least five times higher. A VulnCheck researcher reported that 2 of 300 vulnerabilities discovered by Anthropic or Project Glasswing had been exploited in the wild. The article also reports that AI-generated vulnerability patches can introduce security risks.
CVE
CVE-2026-26980The two vulnerabilities that have witnessed active exploitation efforts are CVE-2026-26980, an SQL injection flaw in Ghost CMS, and CVE-2026-61500, a session forgery flaw in Rejetto HTTP File Server.CVE-2026-61500The two vulnerabilities that have witnessed active exploitation efforts are CVE-2026-26980, an SQL injection flaw in Ghost CMS, and CVE-2026-61500, a session forgery flaw in Rejetto HTTP File Server.
People
Vendors
AnthropicAnthropic on Tuesday said it's expanding a program that allows vetted cybersecurity professionals to test its advanced artificial intelligence (AI) models with reduced safeguards and blocking classifiers, as the companyVeracodeby threat actors or capable of causing significant impacts. Moreover, as demonstrated by 1Password and Veracode, AI-generated vulnerability patches can themselves introduce new security risks of their own.
Products
Claude Mythos 5.1work. Each tier comes with access to its models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models moving forward.Claude Opus 5.5apply for one that best aligns with their work. Each tier comes with access to its models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models moving forward.Claude Sonnet 5.5best aligns with their work. Each tier comes with access to its models, including Claude Opus 5.5, Claude Sonnet 5.5, Claude Mythos 5.1, and new models moving forward.Ghost CMSThe two vulnerabilities that have witnessed active exploitation efforts are CVE-2026-26980, an SQL injection flaw in Ghost CMS, and CVE-2026-61500, a session forgery flaw in Rejetto HTTP File Server.Rejetto HTTP File ServerThe two vulnerabilities that have witnessed active exploitation efforts are CVE-2026-26980, an SQL injection flaw in Ghost CMS, and CVE-2026-61500, a session forgery flaw in Rejetto HTTP File Server.