Choose Rich Live
Tech

Anthropic expands Cyber Verification Program into three tiers with reduced safeguards

Anthropic has revamped its Cyber Verification Program into a three-tier system, folding in Project Glasswing and widening access to its advanced AI models with reduced safeguards for vetted cybersecurity teams and government partners.

The Anthropic logo illuminated on a screen with a keyboard in the foreground under blue lighting.
Image: @analyticsinme

Anthropic has overhauled its Cyber Verification Program into a new three-tier framework, integrating its Project Glasswing initiative into the offering. The updated setup broadens access to the company's frontier artificial intelligence models for vetted cybersecurity specialists, critical infrastructure operators, and university and government researchers.[3][6][11][12][10]

The expansion follows vulnerability research under Project Glasswing, which helped uncover more than 100,000 software flaws this year, according to The Hindu, while commentator Lark Davis placed the count at over 129,000 verified vulnerabilities. In parallel, Bloomberg reported that Anthropic is collaborating with the U.S. government, enabling a select group of organizations to test cutting-edge cyber capabilities on critical systems such as power grids and flight controls.[5][10][9][1]

Under the program, approved participants can test models with fewer safeguards. Analytics Insight reported that the rollout encompasses Claude Mythos 5.1, Opus 5.5, and Sonnet 5.5, noting that Mythos offers relaxed boundaries for approved cybersecurity and life sciences research. While Crypto Briefing noted the tiered approach aims to strengthen defensive capabilities while balancing dual-use technology risks, The New Stack reported that operating with fewer safeguards warrants deeper review, warning that offensive tests leave the real-world impact on everyday defensive work unclear.[7][5][10][8][4]

Key facts

  • Anthropic restructured its Cyber Verification Program into a three-tier system, merging it with Project Glasswing.
  • The program extends access to frontier models, including Claude Mythos 5.1, Claude Opus 5.5, and Claude Sonnet 5.5, to vetted researchers, critical infrastructure operators, and university and government groups.
  • A select group of organizations will evaluate Anthropic's cyber capabilities in collaboration with the U.S. government.
  • The Hindu reported Project Glasswing uncovered more than 100,000 software vulnerabilities this year, while Lark Davis reported at least 129,000 verified vulnerabilities.
  • The program reduces model safeguards across tiers to permit defensive stress-testing on infrastructure such as power grids and flight controls.
  • The New Stack reported that lowering safeguards demands deeper scrutiny and leaves the impact of offensive testing on everyday defensive security unclear.

Sources · 10 sources

  1. CB

    Crypto BriefingArticle ·

    Anthropic partners with US government to test advanced AI cyber capabilities Anthropic's collaboration with the US government could boost its market position and valuation, influencing AI industry dynamics significantly. The post Anthropic partners with US government to test advanced AI cyber capabilities appeared first on Crypto Briefing .

    Open source
  2. SE

    SecurityWeek@SecurityWeekPost on X ·

    Anthropic Introduces 3-Tier Cyber Verification Program for AI Access https://t.co/4b8eUQQ3mx

    Open source
  3. TE

    Techmeme@TechmemePost on X ·

    Anthropic integrates its Cyber Verification Program and Project Glasswing into a new, expanded version of CVP, with three tiers and access to its latest models (Anthropic) (Visit Techmeme dot com for the link and full context!)

    Open source
  4. TN

    The New Stack@thenewstackPost on X ·

    Anthropic expands verified cyber access, but fewer safeguards require deeper review. Offensive tests leave the impact on everyday defensive work unclear. https://t.co/Pky1zPUzdp

    Open source
  5. TH

    The Hindu@the_hinduPost on X ·

    Anthropic is expanding a programme that allows vetted cybersecurity professionals to test its most powerful AI ​models with fewer safeguards, after its Project Glasswing initiative helped ‌uncover more than 100,000 software vulnerabilities this year. https://t.co/mnLcS3pBVz

    Open source
  6. SI

    SiliconANGLE@SiliconANGLEPost on X ·

    Anthropic folds Project Glasswing into an expanded three-tier Cyber Verification Program https://t.co/S8x4aZTXqU

    Open source
  7. AI

    Analytics Insight@analyticsinmePost on X ·

    𝐀𝐧𝐭𝐡𝐫𝐨𝐩𝐢𝐜 𝐎𝐩𝐞𝐧𝐬 𝐀𝐝𝐯𝐚𝐧𝐜𝐞𝐝 𝐂𝐥𝐚𝐮𝐝𝐞 𝐌𝐨𝐝𝐞𝐥𝐬 𝐭𝐨 𝐌𝐨𝐫𝐞 𝐂𝐲𝐛𝐞𝐫𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐅𝐢𝐫𝐦𝐬 Anthropic is giving trusted cybersecurity teams wider access to its latest Claude models, including Opus 5.5, Sonnet 5.5 and Mythos 5.1. Mythos offers fewer limits for approved users working on cybersecurity and life sciences research. #Anthropic #ClaudeAI #Cybersecurity #ArtificialIntelligence #AI #TechNews #analyticsinsight #analyticsinsightmagazine Read More 👇 https://t.co/ntEqtT7uk7

    Open source
  8. CB

    Crypto BriefingArticle ·

    Anthropic expands Cyber Verification Program with new access tiers Anthropic's tiered access expansion could enhance cybersecurity defenses while balancing the risks of dual-use technology proliferation. The post Anthropic expands Cyber Verification Program with new access tiers appeared first on Crypto Briefing .

    Open source
  9. BL

    Bloomberg@businessPost on X ·

    Anthropic is expanding access to its most advanced AI models, allowing a select group of organizations to test the startup’s cutting-edge cyber capabilities in collaboration with the US government https://t.co/whNv3McXh4

    Open source
  10. LD

    Lark Davis@LarkDavisPost on X ·

    Anthropic is opening its frontier models, including Claude Mythos 5.1 and Opus 5.5, to more vetted cybersecurity researchers and critical infrastructure operators. After Project Glasswing turned up at least 129,000 verified vulnerabilities this year, the expanded program cuts safeguards across three tiers so defenders can stress-test systems like power grids and flight controls. Turns out keeping software secure requires the AI to break it first.

    Open source
  11. QU

    Quartz@qzPost on X ·

    Anthropic Opens AI Cyber Access to More Security Firms Anthropic has restructured its Cyber Verification Program into three tiers, giving a wider range of cybersecurity organizations access to its mos https://t.co/L45SkPSta5 #CyberVerification #AnthropicClaude #CybersecurityAI https://t.co/ZamlQBs6GN

    Open source
  12. QU

    Quartz@qzPost on X ·

    Anthropic is giving cybersecurity teams expanded access to its most powerful AI Anthropic restructured its Cyber Verification Program into three tiers, letting a broader range of security orgs — from university researchers to government r #Anthropic #CyberSecurity #ClaudeMythos https://t.co/EMA24bXLqV

    Open source
  13. SE

    SecurityWeek@SecurityWeekPost on X ·

    Anthropic Introduces 3-Tier Cyber Verification Program for AI Access https://t.co/4b8eUQQ3mx

    Open source