#0843
`Project Glasswing` reports 10k+ severe vulnerabilities found with `Mythos Preview`
50radar
Mythos PreviewCybersecurity AI model — built for large-scale vulnerability finding
AI-assisted security work has moved from finding bugs to verifying and patching them. Dependency risk will rise before fixes land, so watch critical packages closely.
- About 50 partners used
Claude Mythos Previewto find 10k+ high or critical vulnerabilities across core software. - Cloudflare found 2,000 bugs, including 400 high or critical, with a false-positive rate it judged better than human testers.
- Disclosure lags discovery by 90 days, or about 45 days after patches. Public CVEs will trail model capability.
- The practical bottleneck is now triage, disclosure, and patch rollout. Automated scans alone are not enough without update discipline.
Source: www.anthropic.com/research/glasswing-initial-updateRead original →