The Pulse
Anthropic Opens Free AI Vulnerability Scans to Open-Source Projects
Anthropic launched OSS Scanner on October 8, offering eligible open-source projects periodic vulnerability scans from its strongest models at no cost. The opt-in reports arrive without human review and can include a proof of concept and a p
AI.info Team · Written with AI ·
Anthropic puts unreviewed findings in maintainers’ hands
Anthropic launched OSS Scanner on October 8, offering eligible open-source projects recurring vulnerability scans from its strongest AI models at no cost. The opt-in service sends reports directly to maintainers without human review, a faster route than the company’s existing process of checking findings before disclosure.
Reports include a self-contained reproducer showing how a bug can be triggered, an explanation and, when available, a candidate patch. Anthropic says they may also include a bisection to identify when a bug was introduced. The company warns that model-generated reports can be wrong, including on severity.
The launch answers a capacity problem Anthropic says it encountered while scanning software under Project Glasswing. Over six months, the company found more than 29,000 candidate vulnerabilities but manually reviewed and triaged about 6,000. It says maintainers increasingly asked to receive the remaining findings, even when they had not been validated.
A faster track, with a different burden
OSS Scanner is aimed at maintainers able to assess incoming reports themselves. Anthropic says projects without the capacity to handle unreviewed findings will continue to receive human-verified disclosures through its coordinated vulnerability disclosure process. The scanner’s submissions do not carry a 90-day disclosure deadline while they remain unvalidated; Anthropic says that period may apply if it later verifies a finding through its standard process.
Core maintainers can apply by submitting a pull request to Anthropic’s enrollment repository. Projects are assessed individually, using criteria similar to Google’s OSS-Fuzz: their impact on infrastructure and user security, exposure to remote attacks and how many users or other projects depend on them. Anthropic says it manually checks that applicants are core maintainers.
Scans recur after enrollment, but the timing may vary with the number of projects in the program and how widely each is used. Maintainers can pause reports or remove a project by submitting another pull request. Anthropic says its scanning agents run in hardened sandboxes with internet access disabled.
Anthropic reports one invalid finding in an early test
To assess an early version, Anthropic’s penetration testers reviewed 97 scanner findings rated critical or high across 48 projects. The company says 85 met the standard for its coordinated disclosure process. Of the other 12, it classified 11 as real but duplicative or otherwise overlapping with findings from the scan; one was invalid.
That result is not the same as a guarantee that future reports will be correct. Anthropic says some maintainers have found inflated severity ratings or cases where the scanner misunderstood a project’s threat model. In its launch post, the company says it expects a true-positive rate above 90%, while acknowledging that reports can contain inaccuracies.
Some maintainers who tested the service praised the reports. “Of the 74 reports we received, all but two were valid, and five became CVEs,” wrote Todd Ouska of wolfSSL. He said attached patches fit the project’s existing process for verifying and fixing issues.
Applications focus on widely relied-on code
Anthropic says the scanner builds on work that scanned hundreds of widely used projects through Project Glasswing, where people reviewed potential vulnerabilities before privately reporting them. OSS Scanner shifts part of that work to a faster, automated channel for projects that choose to receive raw model output. The company says it has already sent nearly 5,000 reports directly to maintainers who requested unvalidated findings.
The free service is supported by Anthropic’s Defender Advantage Fund, alongside funding for open-source organizations and groups that coordinate vulnerability reports. The company also offers qualifying maintainers free Claude Max subscriptions through Claude for Open Source. For OSS Scanner, the concrete trade-off is speed for scrutiny: maintainers get reports sooner, but they—not a human reviewer at Anthropic—must decide what is real and what to fix.