Breaking

Anthropic Expands Vetted Claude Cyber Access After Glasswing Reports 129,000 Vulnerabilities

October 7, 2026 at 07:40 EDT

Security
Foundation Models
AI Agents

Anthropic on Oct. 6 expanded its Cyber Verification Program, giving qualifying security professionals tiered access to Claude models with reduced cyber blocking classifiers while retaining vetting, monitoring and usage controls. The change folds Project Glasswing and the earlier CVP into one program after Anthropic said Glasswing partners verified at least 129,000 software vulnerabilities between April and July 2026.

Oct. 6, 2026 | Anthropic

A vetted lane for cyber Claude

Anthropic merged Project Glasswing and CVP into three access tiers, widening reduced-safeguard cyber work while keeping verification and monitoring controls.

CVP access tiers

129,000+

verified Glasswing flaws

33,000+

critical or high severity

34/50

Red Team tasks completed

Verified vulnerabilities, scaled

129,000+

Glasswing partners

5,500

open-source scans

Each block represents about 5,500 verified vulnerabilities; the partner-reported total is roughly 24 times the open-source scan count.

Access moves through gates

Step 1

Apply and verify security work

Step 2

Receive Defense, Red Team or Specialized Access

Step 3

Validate, disclose and patch findings

Why defenders want it

Less friction for reverse engineering, malware analysis, vulnerability validation and exploit-chain reviews across complex systems.

Why caution remains

Discovery is not remediation: teams still need human review, deduplication, coordinated disclosure and production patches.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.