Project Glasswing: An Initial Update

Project Glasswing, a collaborative initiative launched in May 2026 to identify vulnerabilities in critical software before AI models can exploit them, has discovered over 10,000 high- or critical-severity vulnerabilities in its first month, with partner organizations like Cloudflare finding bugs at rates over ten times higher than previous methods. The AI model used, Claude Mythos Preview, has demonstrated exceptional performance on multiple security benchmarks and real-world testing, though the team is limiting detailed public disclosure of findings to allow time for patches before vulnerabilities can be exploited. The project highlights a fundamental shift in cybersecurity where AI's ability to find vulnerabilities now exceeds the pace at which organizations can verify, disclose, and patch them.