Does Anthropic review OSS Scanner reports before sending them?

Anthropic states that OSS Scanner reports are fully model-generated and sent without human review or triage, so some may be incorrect or invalid. Because no person checks them first, Anthropic does not apply a 90-day disclosure deadline and does not make the findings public. Maintainers receive the report by email and decide what is real.

How accurate are OSS Scanner findings?

Anthropic reports that expert penetration testers reviewed 97 critical and high-severity findings across 48 projects, and that 85 of them, or 88 percent, met the bar for its coordinated vulnerability disclosure process. Of the remaining 12, Anthropic says 11 were real but duplicated known issues and one was a false positive. Anthropic adds that it cannot guarantee the scanner will be perfect.

How does a project enrol in OSS Scanner?

Core maintainers enrol by opening a pull request on the anthropics/oss-scanner repository that adds a project directory holding a project.yaml file and a Dockerfile, with an optional threat_model.md. Anthropic accepts projects case by case and says it looks for critical impact on infrastructure and user security. Anthropic merges only enrolment pull requests and does not accept other contributions.

What did Anthropic find before launching OSS Scanner?

Anthropic says its models found more than 29,000 candidate vulnerabilities over six months, that its team manually reviewed and triaged about 6,000 of them, and that nearly 5,000 reports went directly to maintainers who asked for them. Anthropic cites wolfSSL, which reported that all but two of the 74 reports it received were valid and that five became CVEs.

INFOLab safety

Anthropic OSS Scanner: model-generated bug reports, no human review

OSS Scanner gives enrolled open-source projects free Claude vulnerability scans, with findings emailed to maintainers without human review.

October 8, 2026 · Anthropic · Research by Frontier Red Team

Anthropic

INFOLab safety

Anthropic launched OSS Scanner, a free opt-in service that scans enrolled open-source projects with Claude and emails findings to maintainers. Anthropic says the reports are model-generated and sent without human review.

01What happened

Anthropic's Frontier Red Team launched OSS Scanner on October 8, 2026, an opt-in vulnerability-finding service for open-source projects that Anthropic models on Google's OSS-Fuzz. Anthropic says the service grew out of its Project Glasswing work and runs on its strongest models, including Claude Mythos. The launch landed alongside Anthropic's wider Cyber Mission announcement, which also set up a Critical Infrastructure Defense Program.

02How it works

Enrolment runs through a pull request to the anthropics/oss-scanner repository that adds a project.yaml and a Dockerfile, plus an optional threat model. The scanner then builds the project in an isolated virtual machine and scans it with no internet access. Each report carries a self-contained reproducer, an explanation that bisects to the commit which introduced the bug where possible, and a candidate patch when one exists.

03Who is affected

Core maintainers of open-source projects can apply, and Anthropic says it looks for projects with critical impact on infrastructure and user security, judged case by case on criteria similar to OSS-Fuzz. Reports go to the single contact address named in project.yaml, which is public. Projects can attach a PGP key to encrypt reports, or set a flag to pause them.

04What to do

Maintainers who enrol should plan triage capacity first, because Anthropic states the findings arrive unreviewed. Use a security alias rather than a personal address, add a threat_model.md to steer how severity is rated, and set the disabled flag to true to pause reports when the team cannot keep up.

Questions people ask

Does Anthropic review OSS Scanner reports before sending them?: Anthropic states that OSS Scanner reports are fully model-generated and sent without human review or triage, so some may be incorrect or invalid. Because no person checks them first, Anthropic does not apply a 90-day disclosure deadline and does not make the findings public. Maintainers receive the report by email and decide what is real.
How accurate are OSS Scanner findings?: Anthropic reports that expert penetration testers reviewed 97 critical and high-severity findings across 48 projects, and that 85 of them, or 88 percent, met the bar for its coordinated vulnerability disclosure process. Of the remaining 12, Anthropic says 11 were real but duplicated known issues and one was a false positive. Anthropic adds that it cannot guarantee the scanner will be perfect.
How does a project enrol in OSS Scanner?: Core maintainers enrol by opening a pull request on the anthropics/oss-scanner repository that adds a project directory holding a project.yaml file and a Dockerfile, with an optional threat_model.md. Anthropic accepts projects case by case and says it looks for critical impact on infrastructure and user security. Anthropic merges only enrolment pull requests and does not accept other contributions.
What did Anthropic find before launching OSS Scanner?: Anthropic says its models found more than 29,000 candidate vulnerabilities over six months, that its team manually reviewed and triaged about 6,000 of them, and that nearly 5,000 reports went directly to maintainers who asked for them. Anthropic cites wolfSSL, which reported that all but two of the 74 reports it received were valid and that five became CVEs.

Sources 3 links, all fetched

1anthropic.comFrontier Red Team Post
2anthropic.comAnthropic Cyber Mission
3github.comOSS Scanner Repo