Glasswing Floods Maintainers, Fixes Few Findings
VulnCheck’s seven-month review found 442 CVEs attributed to Anthropic’s Glasswing effort, but only two appear to have been exploited in the wild. Anthropic’s ledger lists 37,966 findings, yet just 217 are marked fixed and 19.4% have been made public.
Glasswing demonstrates that AI can scale vulnerability discovery faster than open-source communities can validate, prioritize, and remediate it. The bottleneck has shifted from finding bugs to producing trustworthy, actionable reports.
- –Only 0.45% of tracked Glasswing-linked CVEs appear in VulnCheck’s exploited-in-the-wild catalog.
- –The gap between Anthropic’s reported true-positive rates and maintainer-level reviews suggests severity inflation and noisy triage.
- –Duplicate and withdrawn findings outnumber confirmed fixes, adding operational cost for already understaffed projects.
- –OSS Scanner’s model-generated fast track may help well-resourced maintainers, but risks overwhelming smaller teams with unverified reports.
- –The program’s real legacy may be proving that vulnerability remediation—not vulnerability discovery—is the harder scaling problem.
DISCOVERED
4h ago
2026-11-11
PUBLISHED
7h ago
2026-11-10
RELEVANCE
AUTHOR
panic