AIMay 23, 2026Updated: May 23, 20266 min read

Anthropic's Project Glasswing: AI Finds 10,000+ Critical Bugs in One Month

Anthropic's Mythos Preview model has found over 10,000 high- or critical-severity vulnerabilities across the world's most critical software in just one month — with an over 90% true positive rate, fundamentally reshaping what AI means for cybersecurity.

L

Lugon

Vibe Engineer

Share article
Anthropic's Project Glasswing: AI Finds 10,000+ Critical Bugs in One Month

The Bug-Finding Revolution Is Here

One month after launching Project Glasswing, Anthropic and roughly 50 partners have used the Mythos Preview model to surface more than 10,000 high- or critical-severity vulnerabilities across the world's most systemically important software.

That's not a typo. The bottleneck in software security has fundamentally shifted — from finding bugs to patching them.

Mythos Preview: Benchmark-Breaking Results

Mythos Preview isn't just incrementally better. Independent testing tells a stark story:

  • UK AI Security Institute: First model to solve both cyber ranges end to end.
  • Mozilla: Found and fixed 271 vulnerabilities in Firefox 150 — over 10x more than Firefox 148 with Claude Opus 4.6.
  • XBOW: Reports a "significant step up over all existing models" with "absolutely unprecedented precision."
  • Cloudflare: Found 2,000 bugs (400 high/critical severity) with a false-positive rate better than human testers.
  • ExploitBench & ExploitGym: Mythos Preview ranks strongest on exploit development capabilities.

The Real Bottleneck: Patching, Not Finding

The industry standard is 90-day coordinated disclosure. With AI generating findings at 10x+ the historical rate, the math breaks down. Cloudflare's latest release had 5x more patches than usual. Microsoft says patch volumes "will continue trending larger for some time."

In open-source, Mythos Preview found an estimated 6,202 high/critical vulnerabilities across 1,000+ projects. Independent firms confirmed a 90.6% true positive rate. One notable find: a certificate forgery vulnerability in wolfSSL (CVE-2026-5194), a library used by billions of devices.

What This Means for Builders

If you maintain or ship software: expect your dependency audit surface to expand dramatically. AI-assisted security tooling is no longer theoretical. Teams that integrate these tools into CI/CD pipelines today will have a structural advantage.

The question is no longer *whether* AI can find your bugs. It's whether you can patch them faster than AI finds them.

Credit

aicybersecurityanthropicglasswingvulnerability
Share article
Start Your Project

Ready to transform?

Discover how TeguFy can help your business simplify, amplify, and fortify with AI, Blockchain, and cutting-edge technology.