Anthropic Launches Free OSS Scanner for Open-Source Vulnerability Audits

Anthropic reported 77 critical and 803 high model-rated findings 11 days after launch, according to the report. As of Oct. 19, 114 projects had been accepted from about 650 applications. The service runs on Anthropic’s infrastructure at no cost to maintainers. Maintainers supply a Dockerfile; the project is built with network access, then audited in an isolated virtual machine without internet access.

The scanner is run jointly by Anthropic’s Frontier Red Team and Safeguards Research team. The design borrows from Google’s OSS-Fuzz but replaces fuzzers with language models, MarkTechPost reported. Anthropic said the program is aimed at a review bottleneck: over six months, its models surfaced more than 29,000 candidate vulnerabilities; human reviewers triaged roughly 6,000, and nearly 5,000 more reports went directly to maintainers who asked for everything, validated or not. OSS Scanner turns that ad hoc arrangement into an opt-in program.

Core maintainers enroll by opening a pull request on anthropics/oss-scanner and adding projects//project.yaml based on a project template. Required fields are the repository, a primary contact and a Dockerfile path. Optional fields include a threat model, PGP key for encrypted reports, auto_ccs, homepage and disabled flag. Agents in the pipeline double-check bugs, run root-cause analysis and propose patches. Each report contains a self-contained reproducer and an explanation, and where possible a bisection to the commit that introduced the bug and a candidate patch. An optional threat_model.md defines adversarial inputs, out-of-scope code and a severity rubric; without it, the scanner guesses, according to the report.

Anthropic’s penetration testers checked 97 critical and high findings across 48 projects. Of those, 85, or 88%, met the bar for Anthropic’s coordinated vulnerability disclosure process, according to the report. Eleven were real but duplicated known issues or other scan findings, and one was invalid. Maintainer data pointed in the same direction. wolfSSL received 74 reports; all but two were valid, and five became CVEs. curl patched CVE-2026-92392, a high-severity HTTP/2 upload use-after-free affecting curl 8.12.0 to 8.22.0 and fixed in 8.23.0, released Oct. 14, 2026.

The absence of human review before delivery is the program’s main trade-off, according to the report. It speeds findings to maintainers but shifts triage work onto project teams. Eleven of the 97 audited critical and high findings duplicated known issues, and maintainers reported inflated severity, MarkTechPost said. The report also compared OSS Scanner with other open-source security efforts: Google OSS-Fuzz uses fuzzing and sanitizers and offers a 90-day disclosure deadline plus a 14-day grace period, while OSS Scanner places no deadline on unvalidated findings. Anthropic links the timing of the launch to benchmark progress; on CyberGym, large language models went from finding under 20% of vulnerabilities to over 85%, the report said. Anthropic’s broader CVD dashboard listed 6,157 disclosed vulnerabilities across 591 projects as of Oct. 2, with 516 patched.