Anthropic Offers Open-Source Projects Free Mythos Bug Scans With No Human Review

Anthropic launched OSS Scanner on Oct. 8, a free opt-in service that sends model-generated vulnerability reports to eligible open-source projects without human review. Its models flagged 29,439 candidate vulnerabilities since Nov. 1, 2025, and external security firms have reviewed 6,123 of them as of Oct. 2, totals covering its wider disclosure program, not OSS Scanner alone. Skipping review, Anthropic says, enables faster and more frequent scanning but “means that it is possible reports will be incorrect or invalid.” The checking work moves to the maintainer.

The review backlog

External security firms had reviewed fewer than a quarter of the candidate vulnerabilities by Oct. 2. Anthropic said it had sent nearly 5,000 unverified reports by the Oct. 8 launch to maintainers who asked for everything it had found, including proposed patches.

The disclosure dashboard recorded 6,157 findings reported to maintainers as of Oct. 2, including 4,824 sent directly by Anthropic that “may contain false positives.” Of the 6,157, 516 had been patched upstream. Human-reviewed disclosures will continue, especially for projects lacking resources to assess raw reports.

Enrollment and reports

OSS Scanner uses Claude models, including Claude Mythos. Reports contain a self-contained reproducer, an explanation of the vulnerability and, where possible, bisection identifying when the bug entered the code. Candidate patches accompany findings when available. Scans run offline in hardened sandboxes; reports arrive by email.

Core maintainers enroll through a pull request to Anthropic's OSS Scanner GitHub repository, supplying configuration and a Dockerfile to build their project. Anthropic checks that applicants are core maintainers. Eligibility follows criteria similar to OSS-Fuzz, emphasizing established projects with a “critical impact on infrastructure and user security.” The repository received 116 pull requests by Oct. 9, the day after launch; no acceptance count has been published.

The accuracy evidence

Before the Oct. 8 launch, the penetration testers who review Anthropic's disclosure findings checked an early scanner version's 97 critical and high-severity findings across 48 projects. Of those, 85 met Anthropic's coordinated-disclosure standard, 11 were real but duplicated known issues or other scan findings, and one was invalid.

In feedback published at launch, wolfSSL's Todd Ouska said all but two of 74 reports received were valid, and five became CVEs. “With patches attached, the reports slotted right into our existing process to verify and fix issues.”

OpenSSL Corporation's Anton Arapov called AI reports from about 18 months earlier “appalling,” but said Anthropic's raw output was “as good and sometimes better than what we get from people.”

Anthropic expects the service's true-positive rate to exceed 90%. The accuracy figures come from Anthropic's own testers and from maintainers Anthropic selected and quoted; no independent audit of the scanner's output has been published. The validation covered only critical and high-severity findings from an early version.

Maintainer capacity

Some maintainers told Anthropic that severity ratings can be inflated or that the scanner misunderstood their project's threat model. The service FAQ acknowledges that many projects are already overwhelmed by reports. It says OSS Scanner is built for projects “already able to keep up with verified high/critical vulnerability reports.”

OSS Scanner's terms apply Anthropic's consumer terms of service, under which inputs and outputs may be used for model training.

In July 2025, curl's Daniel Stenberg wrote that about 20% of curl's 2025 security submissions were AI slop, and by early July about 5% of 2025 submissions had turned out to be genuine vulnerabilities. Each report pulls in three or four of curl's seven security team members for 30 minutes to three hours apiece. In feedback Anthropic published at the Oct. 8 launch, Stenberg said OSS Scanner had found “one of the worst curl vulnerabilities reported in the last few years.”

Anthropic sets no 90-day disclosure deadline for unvalidated OSS Scanner findings. Later human validation can start that clock. The FAQ says Anthropic “may in the future impose a disclosure period on some high-severity vulnerability reports.”

Mistral Previews 1-Trillion-Parameter Large 4, Pitching Open Weights Against Lock-In

Earlier this year, Hugging Face was breached by rogue OpenAI agents. It turned to leading American closed models to help defend itself, only to find their guardrails blocking the work. The guardrails

The Implicator

Nvidia Says Its New Agent Safety Platform Could Have Stopped the Hugging Face Breach

Nvidia launched its Open Agent Safety Platform on Monday and said it could have stopped OpenAI’s agents from breaching Hugging Face in July. The system pairs an open-source runtime that limits agent a

The Implicator

OpenAI and Anthropic Probe Tens of Thousands of Incidents as OpenAI Halts Training

OpenAI, Anthropic and security researchers are investigating tens of thousands of incidents in which advanced models acted beyond intended limits, while OpenAI has paused training of its most capable

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai