---
title: "Expanding the Cyber Verification Program"
description: "Anthropic is expanding its Cyber Verification Program (CVP) to make advanced AI models with reduced safeguards available to qualifying security professionals. The program now offers three access tiers—Defense Access, Red Team Access, and Specialized Access—each with different verification requirements and security controls. Defense Access allows defensive work like incident response and vulnerability analysis, while Red Team Access adds authorized penetration testing. Specialized Access, with the fewest cyber blocks, is reserved for organizations testing high-risk safety systems and is reviewed in collaboration with the US government. Evaluations on CyScenarioBench showed that the Red Team Access tier did not block any tasks and achieved a 68% success rate, equivalent to no safeguards, while the Defense Access tier blocked 46 of 50 tasks."
---

# Expanding the Cyber Verification Program

[Anthropic](https://yomu.fyi/company/anthropic) · Anthropic · Oct 6, 2026

**Type:** Announcement

## Summary

Anthropic is expanding its Cyber Verification Program (CVP) to make advanced AI models with reduced safeguards available to qualifying security professionals. The program now offers three access tiers—Defense Access, Red Team Access, and Specialized Access—each with different verification requirements and security controls. Defense Access allows defensive work like incident response and vulnerability analysis, while Red Team Access adds authorized penetration testing. Specialized Access, with the fewest cyber blocks, is reserved for organizations testing high-risk safety systems and is reviewed in collaboration with the US government. Evaluations on CyScenarioBench showed that the Red Team Access tier did not block any tasks and achieved a 68% success rate, equivalent to no safeguards, while the Defense Access tier blocked 46 of 50 tasks.

## Context

Cybersecurity capabilities are dual-use, so generally available models have conservative safeguards that block most cyber work. For six months, Anthropic ran Project Glasswing and the Cyber Verification Program (CVP) to give trusted organizations access to reduced safeguards. The expanded program integrates these into one offering with three tiers, aiming to give more security organizations access to advanced capabilities.

## Approach / What changed

The program offers three access tiers: Defense Access for defensive work (incident response, vulnerability analysis, reverse-engineering malware), Red Team Access for authorized penetration testing, and Specialized Access for organizations testing high-risk safety systems (e.g., flight operating systems, power grids) with US government collaboration. Each tier has different verification requirements and security controls. Data retention is required for monitoring, with Enterprise Frontier Safeguards (EFS) coming later for eligible organizations.

## Takeaways

- The CVP now has three tiers (Defense, Red Team, Specialized) with different verification requirements and security controls, allowing security teams to apply for appropriate access.
- On CyScenarioBench, the Red Team Access tier on Claude Opus 5.5 did not block any tasks and completed 34 of 50 tasks (68% success rate), equivalent to no safeguards, while Defense Access blocked 46 of 50 tasks.
- Through Project Glasswing, partners uncovered at least 129,000 verified software vulnerabilities between April and July 2026, with over 33,000 rated critical- or high-severity, and the true impact is estimated at least five times higher.

## FAQ

### What are the three access tiers in the Cyber Verification Program?

The three tiers are Defense Access (for defensive work like incident response and vulnerability analysis), Red Team Access (adds authorized penetration testing), and Specialized Access (fewest cyber blocks, for organizations testing high-risk safety systems, reviewed with US government).

### How did the CVP tiers perform on CyScenarioBench evaluations?

Without CVP access, every task was blocked on the first prompt. In Defense Access, 46 of 50 trials were blocked. In Red Team Access, no blocks occurred and Claude Opus 5.5 completed 34 of 50 tasks, equivalent to the model's 67.6% success rate with no safeguards.

### What impact did Project Glasswing have on vulnerability discovery?

Partners uncovered at least 129,000 verified software vulnerabilities between April and July 2026, with more than 33,000 rated critical- or high-severity. The true impact is estimated to be at least five times higher based on partial data from 33 partner reports.

**Tags:** [AI](https://yomu.fyi/topic/ai), [Claude](https://yomu.fyi/topic/claude), [Cybersecurity](https://yomu.fyi/topic/cybersecurity), [LLMs](https://yomu.fyi/topic/llm)

- Source: [Anthropic](https://www.anthropic.com/news/cyber-verification-program)
- Source URL: https://www.anthropic.com/news/cyber-verification-program
- Ingested by Yomu: 2026-10-06T21:00:11.553Z

[Read original post](https://www.anthropic.com/news/cyber-verification-program)