Anthropic has introduced OSS Scanner, a free, opt-in service that performs recurring vulnerability scans for eligible open-source projects. The service uses Anthropic’s most capable models, including Claude Mythos, and is part of the company’s newly announced Anthropic Cyber Mission.

The Cyber Mission also includes a Critical Infrastructure Defense Program with 11 founding partners. Both programs build on Project Glasswing, which Anthropic began in April and later folded into an expanded Cyber Verification Program.

How Anthropic OSS Scanner Works

OSS Scanner is designed to help open-source maintainers receive regular security findings from AI-driven scans. Each report can include:

  • A proof of concept
  • An explanation of the issue
  • A suggested fix, where possible

The model writes the reports, which are sent directly to maintainers without human review. That approach can make disclosures available faster, but Anthropic says some findings may be incorrect, including cases where severity is rated incorrectly.

Anthropic expects more than 90% of the reported findings to be real bugs. In an earlier test, expert penetration testers reviewed 97 critical- and high-severity findings across 48 projects. They approved 85 findings for disclosure, while one of the remaining findings was a false positive.

Open-Source Project Eligibility and Enrollment

The service is open to qualifying projects whose core maintainers choose to participate. To enroll, core maintainers submit a pull request to Anthropic’s GitHub repository.

Eligibility follows the same standard used by Google’s OSS-Fuzz: a project must have a “critical impact on infrastructure and user security.”

Anthropic’s models had identified more than 29,000 possible vulnerabilities over the prior six months, while staff had reviewed about 6,000 of them. Some maintainers asked to receive all findings, including those that had not been reviewed. OSS Scanner was developed in response to that backlog.

Projects that are unable to keep up with unreviewed findings will continue to receive disclosures that have been verified by people.

What Maintainers Receive From AI Security Reports

The value of a vulnerability report can depend on whether it gives maintainers enough detail to understand and act on the finding. OSS Scanner reports may include a working proof of concept and suggested remediation information when available.

Todd Ouska of wolfSSL said that, among 74 reports the project received, all but two were valid, and five became CVEs. Anton Arapov of the OpenSSL Corporation said that a report with a working exploit attached is “basically job done for an engineer.”

Still, the reports do not receive human review before delivery. Maintainers need to account for the possibility of incorrect findings or severity assessments.

Terms for Participating Open-Source Projects

According to reporting on the service, OSS Scanner operates under Anthropic’s consumer terms. Those terms permit inputs and outputs to be used for model training.

Projects that prefer not to manage raw AI-generated findings can continue receiving human-verified disclosures instead.

Anthropic’s Critical Infrastructure Defense Program

Anthropic’s Critical Infrastructure Defense Program provides Claude models, on-site engineers, and threat research to security providers working with:

  • Power grids
  • Water systems
  • Transportation networks

The program’s founding partners are:

  • Accenture
  • Booz Allen
  • CrowdStrike
  • Deloitte
  • Dragos
  • Hitachi
  • Insane Cyber
  • Nozomi Networks
  • Palo Alto Networks
  • PwC
  • Rockwell Automation

Andrew Turner of Booz Allen described operational technology as “the next frontier for autonomous AI-enabled attacks.” Anthropic has not said whether participating partners will receive free model access or who will cover computing costs.

From Project Glasswing to the Cyber Verification Program

The OSS Scanner and Critical Infrastructure Defense Program build on Project Glasswing. Anthropic started that effort in April, then incorporated it into an expanded Cyber Verification Program on October 6.

Anthropic said Glasswing identified many vulnerabilities, but it has not yet reduced cyber risk enough. The company predicts that AI will favor defenders within two years, while also acknowledging that this may not be true in the near term.

FAQ

What is Anthropic OSS Scanner?

OSS Scanner is a free, opt-in service that runs regular AI-based vulnerability scans for eligible open-source projects. It sends model-written reports to maintainers without human review.

What does an OSS Scanner vulnerability report include?

Reports include a proof of concept, an explanation, and, where possible, a suggested fix. Anthropic notes that some reports may be incorrect or include an incorrect severity rating.

How can an open-source project enroll in OSS Scanner?

A project’s core maintainers can submit a pull request to Anthropic’s GitHub repository. Eligibility follows the OSS-Fuzz standard of having a critical impact on infrastructure and user security.