Anthropic's Red Team cyber tier blocked none of 50 trials in its own test
Anthropic has folded Project Glasswing and its Cyber Verification Program into one programme with three access tiers, and published its own measurement of what each tier stops. In the Red Team Access tier, Anthropic says, "no blocks occurred" when it ran Claude Opus 5.5 through 50 trials of an offensive security evaluation.

A tall comparison table with green ticks spreading further down each column from left to right. Image: Anthropic
The announcement is dated October 6, and the figures come from five attempts at each of 10 CyScenarioBench challenges in every tier. Without CVP access, Anthropic reports, "every task was blocked on the first prompt". In Defense Access, "46 of the 50 trials were blocked at some point in the challenge, while the remaining four tasks succeeded".
The upper tiers scored like a model with nothing switched on
In Red Team Access, Anthropic says Claude Opus 5.5 "successfully completed 34 of the 50 tasks", a result it calls "effectively equivalent to the model's 67.6% success rate on this evaluation with no safeguards applied". The company marks that run as "representative of Specialized Access", the tier above.
Anthropic describes CyScenarioBench as "an evaluation that measures whether models can plan and execute multi-stage cyber operations under realistic constraints", and sets out what it expected beforehand: heavy blocking on the generally available model and in Defense Access, "while experiencing no blocks in the Red Team Access and Specialized Access tiers". The evaluation, the model, the safeguards and the scoring are all Anthropic's, and no outside party is named as having run or checked any of it. The figures above were read from the company's post at 22:54 UTC on October 6.
Anthropic sets out what each tier allows
Defense Access "is for defensive work, including security operations center and incident response tasks, reverse-engineering malware, and analyzing and validating vulnerabilities", and Anthropic aims to answer those applications "within a few days".
Red Team Access "adds authorized penetration testing and red-teaming to the defensive uses above", takes "a few weeks" to review, and is "for organizations only; individual researchers are not eligible". People working in it "will still experience real-time blocks on actions that could cause physical harm or mass disruption, such as deploying ransomware, damaging physical systems, or pen testing high-risk safety systems".
Specialized Access "has the fewest cyber blocks" and is "reserved for a limited set of verified organizations that are authorized to test safety systems that could impact people's lives or disrupt markets, such as flight operating systems, power grids, telecom networks, interbank transfer infrastructure, and government administrative networks".
The US government helps review the top tier
For that tier, Anthropic says it currently reviews "every organization in depth in collaboration with the US government". Members of Project Glasswing, which for six months gave a group of organisations access to Claude Mythos, move across with no reapproval for current models.
Data retention is required throughout the programme so Anthropic "can monitor for cyber misuse", with Enterprise Frontier Safeguards promised "later this fall" for organisations that want their data in infrastructure they control. CVP runs on the Claude Platform, Google Cloud's Vertex AI and Microsoft Foundry, and on Amazon Bedrock only for customers eligible for that newer product.
Anthropic used the same shape in biology in September, when it loosened Claude's biology safeguards for vetted research teams. Google took a comparable route on October 2, saying it would release Gemini 4 Argon without cyber guardrails to trusted defenders.
Every applicant goes through one form, and Anthropic says it will verify each one and ask for proof of the security controls the tier requires. Existing CVP members keep their settings for previous models and are evaluated automatically for Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1.