Anthropic opens three Claude access tiers to cyber defenders

For most Claude users nothing changes: the automatic blocks on offensive or ambiguous cybersecurity requests stay on. The change is for organizations that do security work. Anthropic announced on Oct. 6 that its Cyber Verification Program, an application process that eases those blocks for users who show a legitimate need, now has three tiers.

Defense Access is the tier for defenders: monitoring in a SOC (security operations center), incident handling, malware analysis through reverse engineering and vulnerability research. Red Team Access adds penetration tests run with the system owner's permission and red teaming, which means simulating real attacks to measure defenses. For now it goes only to organizations, not individuals. The third, Specialized Access, is reserved for a few verified organizations that evaluate systems critical to lives or markets, such as power grids, telecommunications or flight systems.

Review times and models

According to the company, Defense is reviewed in a few days, Red Team in a few weeks, and Specialized gets an in-depth review done together with the U.S. government. Current program members will be evaluated automatically for new models. It covers Claude Opus 5.5, Claude Sonnet 5.5 and Claude Mythos 5.1, and is offered on three platforms: Claude Platform, Google Cloud's Vertex AI and Microsoft Foundry. Amazon Bedrock has it too, but only for eligible Enterprise Frontier Safeguards customers.

Enrolling requires accepting data retention. Zero retention will arrive "later this fall" with Enterprise Frontier Safeguards.

The company's figures

Anthropic tested the scheme with Claude Opus 5.5 on CyScenarioBench, an internal evaluation of five attempts at each of 10 challenges per tier. Without program access, every task was blocked on the first request. In Defense Access, 46 of 50 attempts were blocked at some point. In Red Team Access there were no blocks and the model completed 34 of 50 tasks, or 67.6 %, which the company says matches its success rate without safeguards.

These numbers have no independent replication, and the 10 challenges are not public. Anthropic adds that Red Team users will still face real-time blocks on actions that cause physical harm or mass disruption. It also cites its Project Glasswing: more than 129 000 verified vulnerabilities between April and July at partner organizations, and more than 33 000 rated critical or high.

What it does not say

The announcement mentions no price and no regions. It does not say whether organizations in Mexico or Latin America can qualify for Red Team or Specialized, and it gives no start date for each tier. Other labs are heading the same way: Google opened Gemini 4 Argon only to cybersecurity defenders on Oct. 1.