www.securityweek.com 7 Oct 2026, 10:07 UTC

Anthropic expands cyber AI access with strict new security tiers

Anthropic expands cyber AI access with strict new security tiers

ANTHROPIC has announced a revamped Cyber Verification Program (CVP) that combines the previous CVP and Project Glasswing into a single, three-tier offering for access to its top AI models. The expanded access includes Claude Opus 5.5, Sonnet 5.5 and Fable 5.1, along with Mythos 5.1 in future models. All tiers carry the same foundational safeguards, with each level imposing its own verification requirements and security controls. The aim, the company says, is to balance the defensive benefits of powerful AI with the risk that similar capabilities could be misused by attackers.

The three tiers are Defence Access, Red Team Access, and Specialized Access. Defence Access covers SOC and incident response work, malware reverse engineering and vulnerability analysis, with eligibility including security teams defending their own systems, critical infrastructure operators, small security firms, open source maintainers and individual researchers with vulnerability histories.

Red Team Access adds authorised penetration testing and red-teaming, but cannot test in ways that could cause mass disruption or physical harm; at present, individuals are not eligible and reviews are anticipated to take a few weeks. Specialized Access is the most restricted tier for testing safety-critical systems such as power grids and interbank transfer networks, with vetting conducted in collaboration with the US government and existing Glasswing members moving into this tier.

Security considerations include data retention so Anthropic can monitor for misuse. Glasswing partners reported at least 129,000 verified vulnerabilities between April and July, with a further 5,500 found by Anthropic’s open source scanning between April and October; more than 33,000 were rated critical or high. Enterprise Frontier Safeguards will later allow eligible customers to store data in their own cloud, while zero data retention remains possible for some models and access paths.

CVP remains available on Claude Platform, Google Cloud Vertex AI and Microsoft Foundry, with Amazon Bedrock access limited to customers eligible for the new safeguards.

View full article

Article by CyberSIXT

Timeline Coverage

Swipe to explore timeline