Anthropic Gives Open Source Maintainers a Fast Track to Mythos Vulnerability Findings

Anthropic launched OSS Scanner, a free opt-in service that uses its Claude models, including Mythos, to find vulnerabilities in open source projects and send maintainers reports with reproduction steps and proposed fixes without waiting for researcher validation. Anthropic said it found more than 29,000 candidate vulnerabilities over six months but manually reviewed and triaged approximately 6,000, and has already sent nearly 5,000 reports to maintainers who requested unvalidated findings. In an early test, penetration testers examined 97 high- and critical-severity findings across 48 projects, of which 85 met Anthropic's coordinated disclosure criteria, 11 were real but duplicated known issues, and one was invalid.

Anthropic Gives Open Source Maintainers a Fast Track to Mythos Vulnerability Findings Anthropic’s free OSS Scanner uses Mythos and other Claude models to find open source vulnerabilities and send maintainers reports with proposed fixes. - Sarah Gooding Anthropic has launched OSS Scanner https://www.anthropic.com/research/launching-opt-in-vuln-finding-service-for-open-source , a free, opt-in service that uses its strongest Claude models, including Mythos, to look for vulnerabilities in open source projects. Participating maintainers receive reports showing how to reproduce the vulnerabilities, along with proposed fixes when available, without waiting for Anthropic’s researchers to review the findings. The reviewers have struggled to keep up with the volume of reports. Over six months, the company says it found more than 29,000 candidate vulnerabilities but manually reviewed and triaged approximately 6,000. Some maintainers who received initial reports asked Anthropic to send the remaining findings for their projects without waiting for validation. The company has already sent nearly 5,000 reports in response to those requests. OSS Scanner lets more open source projects sign up to receive findings without waiting for Anthropic’s researchers to validate them. The goal is to help open source projects turn AI findings into fixes. In its Cyber Mission announcement https://www.anthropic.com/news/anthropic-cyber-mission , Anthropic makes an admission: Project Glasswing uncovered vulnerabilities, but “we haven’t yet achieved a sufficient reduction in cyber risk.” The company says it often saw months pass between discovery and a fix, a remediation bottleneck that has drawn criticism https://www.vulncheck.com/blog/anthropic-glasswing-receipts from security researchers. OSS Scanner is the first step in a larger long-term effort to automate vulnerability triage, develop patches, and research ways to harden or rewrite code. Although the OSS Scanner makes unreviewed findings available to more projects, the extended FAQ https://red.anthropic.com/oss-scanner/ explicitly targets those that can already keep up with verified high- and critical-severity reports and have capacity to investigate more. 85 of 97 evaluated findings met Anthropic’s disclosure criteria 85-of-97-evaluated-findings-met-Anthropic's-disclosure-criteria Anthropic tested an early version by having penetration testers examine 97 high- and critical-severity findings across 48 projects. Of those, 85 met its coordinated disclosure criteria. Eleven more were real issues that duplicated known problems or other findings from the scan. One was invalid. The evaluation does not establish a false-positive rate for every severity level or every project that joins the service. Anthropic also acknowledges that maintainers have challenged inflated severity ratings and findings based on a misunderstanding of their threat models. The distinction between a real bug and an actionable security report also appears in Anthropic’s broader vulnerability disclosure dashboard https://red.anthropic.com/2026/cvd/ . Its true-positive metric includes duplicates and issues maintainers may decline to fix, including findings outside a project’s threat model or in code that is not typically reachable. The dashboard itself describes patch counts as a more reliable, though delayed, measure of impact. PostgreSQL’s Noah Misch said some proposed patches were usable almost unchanged and that early access helped the project resolve issues before a general-availability release. wolfSSL’s Todd Ouska reported that 72 of 74 findings were valid, with five becoming CVEs. Open source projects are already shipping fixes from AI audits Open-source-projects-are-already-shipping-fixes-from-AI-audits Mozilla’s experience shows what these tools can deliver when a project has the capacity to act on their findings. In April, the Firefox team shipped fixes for 271 vulnerabilities https://blog.mozilla.org/en/firefox/privacy-security/ai-security-zero-day-vulnerabilities/ identified during an early Mythos Preview evaluation. Getting there took sustained engineering work. Mozilla’s Bobby Holley described a team working around the clock and warned that other projects might need to reprioritize everything else to tackle the findings. For projects already using AI security tools, each new scan also builds on earlier cleanup. In May, curl maintainer Daniel Stenberg reported that several AI tools had prompted roughly 200–300 bug fixes https://daniel.haxx.se/blog/2026/05/11/mythos-finds-a-curl-vulnerability/ over the preceding eight to ten months, before curl received its first Mythos report. Those tools included AISLE, Zeropath, and OpenAI’s Codex Security. Stenberg cautioned against judging tools by the number of bugs they uncover: earlier scans had more easy targets, while later ones examined code that had already been improved. A smaller haul can be evidence that previous security work paid off. OSS Scanner gives eligible projects a way to make those audits recurring, with Anthropic covering the cost. Maintainers supply the build environment and security context Maintainers-supply-the-build-environment-and-security-context Established projects with significant infrastructure or security impact can apply through the GitHub contribution process https://github.com/anthropics/oss-scanner/blob/main/CONTRIBUTING.md . Maintainers provide a project configuration and a Dockerfile that prepares the code for an offline audit. Anthropic verifies maintainer status, and projects can pause reports or withdraw. Maintainers also get a say in what the scanner goes looking for. An optional threat model.md file spells out attack surfaces, exclusions, severity criteria, and expectations for reports and patches. That context can help prevent the model from treating behavior the project explicitly permits as a security flaw. There is no 90-day clock ticking on every unverified report https://red.anthropic.com/oss-scanner/ what-is-the-public-disclosure-policy . If Anthropic later validates a finding through its human-reviewed disclosure program, the 90-day period starts when the maintainer is notified. Anthropic may change the scanner’s disclosure policy with advance notice and an opportunity to opt out. Better reports still require maintainers to ship fixes Better-reports-still-require-maintainers-to-ship-fixes curl maintainer Daniel Stenberg, whose endorsement appears in the launch, has documented both sides of AI vulnerability reporting. In July 2025, he described low-quality reports https://daniel.haxx.se/blog/2025/07/14/death-by-a-thousand-slops/ consuming hours from multiple security team members. By April 2026, his concern had changed: reports were arriving more frequently and were much better. In “High-Quality Chaos,” https://daniel.haxx.se/blog/2026/04/22/high-quality-chaos/ Stenberg wrote that increased discovery would worsen maintainer overload without more people to handle the work. He also pointed to the next delay: users still have to install the fixed versions. OSS Scanner gives open source projects access to recurring security audits at no cost, helping maintainers uncover flaws that might otherwise go undetected. Reproducers and proposed patches can help them verify and address those findings. Anthropic is also continuing its human-reviewed disclosure process, particularly for projects without resources to triage raw findings. To help maintainers handle the workload that comes with pumping out vulnerability reports, Anthropic is also funding organizations that support open source security https://www.anthropic.com/news/anthropic-cyber-mission , including the Python Software Foundation, the Apache Software Foundation, and Alpha-Omega and OpenSSF through the Linux Foundation. It is also supporting Akrites and Gold Eagle, which collect and coordinate reports from multiple sources to avoid overwhelming maintainers. AI-assisted audits are forcing open source projects to revisit code that years of testing and review had given them reason to trust. Some are clearing out longstanding flaws and others are struggling to absorb the reports. OSS Scanner makes that work available to more projects, but it also exposes how unevenly equipped they are to take it on. The same discovery tools can produce a patch sprint in a well-resourced project and a growing backlog in one maintained by a handful of volunteers. That difference will shape which parts of the software supply chain get safer first and where attackers still have room to work.