Anthropic’s Glasswing Faces Scrutiny Over Finding Quality and Maintainer Burden
Vulnerability Disclosure · 4 sources · Updated 13h ago · First seen Nov 16, 2026

Anthropic’s Project Glasswing vulnerability-discovery initiative is facing scrutiny over the quality of its findings and the validation workload imposed on open-source maintainers. Analyses attributed to VulnCheck and Davi Ottenheimer cited 37,966 dashboard findings but questioned whether Anthropic’s advertised 94.2% true-positive rate reflected broader results; Ottenheimer reported that only 17% of findings had received human review. Feedback from curl maintainer Daniel Stenberg showed that, among 34 published reports, 19 were rejected, six were confirmed as low severity, and nine remained under assessment. The assessments also acknowledged a reported high-severity curl discovery, indicating that concerns about reporting volume do not mean every finding lacks value.
The accounts described substantial gaps between discovery, disclosure, and remediation, although their figures were inconsistent: VulnCheck’s assessment cited 217 findings marked fixed and 7,349 made public, while Ottenheimer’s cited 629 patched findings. VulnCheck reportedly identified 442 Anthropic-attributable CVEs and named CVE-2026-61500 and CVE-2026-26980 as exploited in the wild, but the supplied reporting did not independently substantiate those exploitation claims. For security leaders, the findings underscore the need to distinguish scanner output from validated vulnerabilities and prioritize confirmed impact, affected deployments, exploitation evidence, and available fixes rather than raw discovery totals.


