Claude Found 129,000 Vulnerabilities in Three Months
Anthropic has expanded access to Claude for security teams, with fewer restrictions. The earlier version of the program found almost 129,000 vulnerabilities in three months.
Anthropic, the company behind the Claude AI models, has expanded its Cyber Verification Program. This program gives vetted security teams access to Claude with fewer safety restrictions than the public version has. The access is meant mainly for vulnerability research and pentesting. Pentesting means deliberately attacking systems to find weak spots. Regular code review was already available before, without any special permissions. The earlier version of the program was called Project Glasswing. From April to July 2026, partner security teams and researchers used Claude under Glasswing. They found at least 129,000 confirmed vulnerabilities. More than 33,000 of them were rated high severity or critical. Anthropic says the real-world impact is likely at least five times higher than that number.
These vulnerabilities were found by human security teams and researchers who used Claude as a tool, not by Claude working on its own. Anthropic does not open this access to just anyone. The program has three tiers, and each organization goes through vetting first. For critical infrastructure, things like power grids, flight control systems, or banking systems, Anthropic checks every applicant together with the U.S. government. A haul of confirmed findings this large in just three months shows that agent-assisted security work can already deliver results at scale. But access stays under tight, tiered control, not open to the public.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →▌ Related guides

Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles