“Project Glasswing has enabled researchers to find 37,966” candidate vulnerabilities — yet, seven months after launch, only 217 have been publicly fixed, according to vulnerability researcher Patrick Garrity.
The numbers: dashboard tallies and the narrow slice that’s fixed
Garrity’s November 9 update synthesizes data from Anthropic’s coordinated vulnerability disclosure dashboard (ledger dated November 2, 2026). That ledger lists 37,966 findings attributed to Project Glasswing, of which 7,349 (19.4%) appear in Anthropic’s public tracking register.
Within the 7,349 ledger entries, Garrity counts 5,276 validated as unique, real vulnerabilities; the remainder are recorded as 1,654 duplicates and 419 withdrawn. Of the 5,276 validated entries, 4,866 have been reported to maintainers and 193 are still in “pre-disclosure” status. Garrity’s review finds that just 217 vulnerabilities — under 0.5% of the total 37,966 claimed findings — are marked as fixed.
Anthropic’s operational shift: OSS Scanner and bulk reporting
Anthropic acknowledged on October 8 that it had identified more than 29,000 candidate vulnerabilities but had manually reviewed only about 6,000, saying it remained “bottlenecked on our human capacity to validate these findings.” To address the backlog, the company launched an Open Source Security Scanner (OSS Scanner), described as a self-service tool inspired by Google’s OSS-Fuzz project that lets maintainers request scans and triage findings themselves.
Anthropic also told Garrity it had already sent nearly 5,000 unverified vulnerability reports directly to maintainers who requested bulk delivery. That change shifts more of the triage task from Anthropic’s internal staff to third-party maintainers.
Your scanner finds 4,000 vulns. Which 12 matter?
Nubivance is a Rapid7 Registered Partner delivering vulnerability management as a service - scanning, risk-based prioritization, and remediation follow-through across IT and OT.
Fix the backlogMaintainer experience: the cURL example
Daniel Stenberg, maintainer of the Client and URL (cURL) project, said on November 6 that he received 45 vulnerability reports from Anthropic through Glasswing. After analysing 34 of them using Anthropic’s OSS Scanner and other tools, Stenberg reported that 19 were not vulnerabilities at all and six were confirmed as low severity.
Stenberg’s experience illustrates the practical triage burden now facing maintainers who accept bulk, unverified findings: parsing duplicates, dismissing false positives, and assigning severity — work Anthropic originally framed Glasswing as helping to reduce.
Garrity’s critique: volume without remediation
Garrity’s November 9 report characterizes Project Glasswing’s output as “overwhelming” maintainers with “underwhelming findings.” He notes that since his prior analysis in September, Anthropic added 4,613 findings to the ledger while only 15 additional items were marked as fixed in the same interval; 1,668 of the new ledger additions were marked as duplicates.
Garrity also questioned Anthropic’s asserted true positive rate of 94.2%, writing it “raises questions about how it accounts for duplicates and withdrawn findings.” He concluded: “AI is a valuable tool for finding vulnerabilities, and I'm not discounting that. But seven months in, the evidence still doesn't match the launch narrative. The CVEs are a sliver of this year's total, almost none are being exploited, fixes have stalled while duplicates and withdrawals pile up, and the triage burden is shifting onto the open source maintainers Glasswing promised to help. The receipts keep coming in. They still don't reconcile.”
Infosecurity contacted Anthropic for comment, but the company had not responded by the time of publication.
What this means for open-source maintainers, Anthropic, and security teams
- Open-source maintainers: Those who opt into Anthropic’s bulk deliveries or OSS Scanner face increased triage work — confirming whether findings are true, deduplicating reports, and reclassifying severity. The cURL maintainer’s analysis (19 of 34 reports not vulnerabilities) is an early, concrete data point.
- Anthropic: The company has signaled it is attempting to scale disclosure by delegating triage (OSS Scanner) and delivering unverified reports, but Garrity’s tally suggests fixes and reconciled accounting have not kept pace with reported findings. Questions remain about how duplicates and withdrawals are handled in high-level metrics like the cited 94.2% true positive rate.
- Security teams and consumers of open-source code: For organizations relying on disclosed fixes, the data shows a small fraction of Glasswing-identified candidates have reached remediation. Teams should treat bulk AI-derived reports as requiring validation rather than immediate actionable fixes.
The record presented by Anthropic’s ledger and Garrity’s audit points to a clear mismatch: a large, automated discovery pipeline generating tens of thousands of candidates, and a far smaller human-driven closure rate. Whether Anthropic’s OSS Scanner and maintainers’ willingness to accept bulk reports will accelerate remediation, or instead simply move the bottleneck to open-source projects, is the concrete question the ledger data leaves on the table.
