Skip to content
sobolewski.proAI today, every eveningSubscribe

Once a day. Unsubscribe with one click. Privacy policy

Cyber Verification Program: Red Team on Enterprise and API only

The Cyber Verification Program after the November 18 update: Red Team Access and Specialized Access only on Enterprise and API plans. What about Team?

4 min read

This article in other languages:Polski

In this article
  1. What changed in the Cyber Verification Program?
  2. What about the Team plan?
  3. What do the three tiers look like after the change?
  4. How much does a higher tier give you?
  5. FAQ

On November 18, 2026, Anthropic updated its Cyber Verification Program announcement. A note at the bottom of the page says plan availability changed for the Red Team Access and Specialized Access tiers, and that the criteria for the latter were clarified.

What changed in the Cyber Verification Program?

The program has run in its current form since October 6. Verified security professionals get models with less restrictive cyber classifiers, at three access tiers. I covered the expansion itself when writing about the vulnerabilities Claude found.

The updated text says Specialized Access, like Red Team Access, is available only on Enterprise and API plans. The second clarification is about who qualifies: operating critical infrastructure is not enough for Specialized Access. You have to show that you are authorized to test that infrastructure, and describe work that can’t be done under Red Team Access.

Anthropic doesn’t say what the earlier version read, so the exact scope of the change can’t be worked out from the announcement.

What about the Team plan?

Here two official sources disagree. On the day I’m writing this (November 19), the Help Center article still lists “Team, Enterprise, and API” plans for Red Team Access and Specialized Access. Its availability table says access on a Team plan covers every user, which is why Red Team Access is capped at 75 seats and Specialized Access at 25.

The updated announcement no longer mentions Team. The simplest explanation is that Team was dropped from the two higher tiers and the Help Center hasn’t caught up, but that is my inference from comparing the two pages, not confirmed information. Teams on the Team plan that were planning to apply for Red Team Access should check the application form before submitting.

What do the three tiers look like after the change?

Defense Access Red Team Access Specialized Access
What for SOC, incident response, malware analysis, vulnerability validation plus authorized penetration testing and red teaming red teaming of systemically important infrastructure (power grids, telecommunications, interbank settlement)
Who organizations and individuals organizations only a limited group of verified organizations
Plans per the announcement not stated Enterprise and API Enterprise and API
Review time a few days a few weeks longer, each application is reviewed together with the US government

Every tier covers Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, and, according to Anthropic, future models too. The Help Center adds that Defense Access is available on all paid plans (Pro, Max, Ultra, Team, Enterprise) and through the API, and that the program isn’t offered on the free plan.

So nothing changes for individuals. A researcher with a track record of reported vulnerabilities can still apply for Defense Access, and Red Team Access was for organizations only from the start.

How much does a higher tier give you?

The gap between tiers shows best in a test Anthropic described in the same announcement. Claude Opus 5.5 was given 10 tasks from CyScenarioBench (multi-stage offensive operations), with five attempts at each:

  1. without program access, every task was blocked at the first prompt,
  2. under Defense Access, 46 of 50 attempts were blocked,
  3. under Red Team Access nothing was blocked, and the model completed 34 of 50 tasks, the same number as with no safeguards at all.

Moving from Defense to Red Team therefore means going from “almost everything blocked” to “nothing blocked”. Hence the longer vetting and, as this update shows, a shrinking list of plans. Even Red Team Access keeps blocks on actions that could cause physical harm or mass disruption, such as deploying ransomware.

There is one more date for people on Defense Access. From December 15, 2026, the Help Center says this tier will require phishing-resistant MFA and will stop accepting static API keys (Anthropic recommends moving to Workload Identity Federation now). I described how to keep keys under control day to day in the guide to AI agent security.

FAQ

Can an individual join the Cyber Verification Program?

Yes, but only at the Defense Access tier. Among those who qualify, the announcement lists open source maintainers and researchers with a track record of reported vulnerabilities.

Which models does the program cover?

Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, plus new models in the future. I wrote about Mythos 5.2 when Claude Fable 5.2 launched.

Is Red Team Access available on the Team plan?

According to the announcement updated on November 18, no: it lists only Enterprise and API plans. On November 19 the Help Center still shows the Team plan with a 75-seat cap, so confirm it in the form before applying.

  • AI
  • Anthropic
  • Claude
  • security

AI today, every evening

The day's biggest AI news and new posts from the blog, every evening. Short, with sources.

Once a day. Unsubscribe with one click. Privacy policy