BRIEFLY.
Claude Mythos AI Model Finds 10,000+ Software Vulnerabilities in a Month
•
1 min read
•
Briefly Editorial Team

Claude Mythos AI Model Finds 10,000+ Software Vulnerabilities in a Month

TL;DR

  • •Claude Mythos identified 10,000+ software vulnerabilities in one month
  • •Cloudflare found 2,000 vulnerabilities, including 400 critical
  • •Anthropic delays public release due to abuse risks

Why it matters

AI models are becoming critical for cybersecurity, but their public deployment requires balancing innovation with safety.

Technical Details

Anthropic launched Project Glasswing in April 2026 using the unreleased Claude Mythos AI. Within a month, it helped partners identify over 10,000 vulnerabilities, including 6,200 critical issues in open-source projects.

Context and Background

Cloudflare discovered 2,000 vulnerabilities, 400 of which were high-risk. Mozilla fixed 271 Firefox bugs, while Microsoft linked increased patch volumes to Mythos analysis. Independent researchers claimed macOS exploits using the model.

Industry Impact

Anthropic is expanding Project Glasswing collaborations with Apple, Google, and Nvidia. The public release remains delayed due to abuse risks, raising questions about AI regulation in cybersecurity.