Anthropic Launches Three-Tier Cyber Verification Program for AI Model Access
Summary
Anthropic combined its Cyber Verification Program and Project Glasswing into three access tiers for AI models, with different verification requirements and safeguards for defensive work, authorized red teaming, and safety-critical testing.
Key points
- The three tiers—Defense Access, Red Team Access, and Specialized Access—cover progressively broader cybersecurity work and have different verification requirements.
- Defense Access supports SOC and incident response work, malware reverse engineering, and vulnerability analysis; eligible applicants include security teams, infrastructure operators, open-source maintainers, and some individual researchers.
- Red Team Access permits authorized penetration testing and red teaming. Actions such as deploying ransomware remain blocked; the tier is currently limited to organizations.
- Specialized Access is reserved for a small number of organizations testing safety-critical systems and is vetted with the US government.
- Anthropic reports that Glasswing partners verified at least 129,000 vulnerabilities from April to July, while its own open-source scanning found 5,500 more; over 33,000 were rated critical or high severity.
- Program participants must accept data retention for misuse monitoring, though some users with zero-retention access can use the program with zero data retention.
- CVP is available through the Claude Platform, Google Cloud Vertex AI, and Microsoft Foundry; Amazon Bedrock access is limited to customers eligible for Enterprise Frontier Safeguards.
Article Details
- Event Type
- Anthropic announced a three-tier Cyber Verification Program that consolidates CVP and Project Glasswing and provides verified users with different levels of access to its AI models.
- Impact
- The program is intended to enable approved organizations and researchers to conduct defensive security work, authorized red teaming, and specialized testing of safety-critical systems, while retaining safeguards against harmful actions. Glasswing partners reported finding at least 129,000 verified vulnerabilities, and Anthropic reported finding another 5,500 through its open-source scanning; more than 33,000 were rated critical or high severity. Anthropic said the figures cover only a subset of participants and estimated the true impact could be at least five times higher.
Vendors
AmazonOn Amazon Bedrock, it’s only available to customers eligible for Enterprise Frontier Safeguards.AnthropicAnthropic on Tuesday announced a revamped Cyber Verification Program (CVP), integrating the CVP and Project Glasswing into a single offering, with three levels of access to its most capable AI models.GoogleCVP is available on the Claude Platform, Google Cloud’s Vertex AI and Microsoft Foundry.MicrosoftCVP is available on the Claude Platform, Google Cloud’s Vertex AI and Microsoft Foundry.
Products
Amazon BedrockOn Amazon Bedrock, it’s only available to customers eligible for Enterprise Frontier Safeguards.Claude MythosGlasswing ran as separate programs: Glasswing gave organizations securing critical software access to Claude Mythos, while the original CVP loosened safeguards on Opus and Sonnet models for approved teams.Claude Opus 5.5Anthropic’s generally available models, including Claude Opus 5.5, Sonnet 5.5 and Fable 5.1, ship with cyber safeguards that block most cyber work.Claude PlatformCVP is available on the Claude Platform, Google Cloud’s Vertex AI and Microsoft Foundry.Cyber Verification Program (CVP)Anthropic on Tuesday announced a revamped Cyber Verification Program (CVP), integrating the CVP and Project Glasswing into a single offering, with three levels of access to its most capable AI models.Enterprise Frontier SafeguardsLater this fall, Enterprise Frontier Safeguards will let eligible customers store data in cloud infrastructure they control.Fable 5.1Anthropic’s generally available models, including Claude Opus 5.5, Sonnet 5.5 and Fable 5.1, ship with cyber safeguards that block most cyber work.Microsoft FoundryCVP is available on the Claude Platform, Google Cloud’s Vertex AI and Microsoft Foundry.Mythos 5.1Under the new structure, all three tiers include Opus 5.5, Sonnet 5.5, Mythos 5.1 and future models.Project GlasswingAnthropic on Tuesday announced a revamped Cyber Verification Program (CVP), integrating the CVP and Project Glasswing into a single offering, with three levels of access to its most capable AI models.Sonnet 5.5Anthropic’s generally available models, including Claude Opus 5.5, Sonnet 5.5 and Fable 5.1, ship with cyber safeguards that block most cyber work.Vertex AICVP is available on the Claude Platform, Google Cloud’s Vertex AI and Microsoft Foundry.