Anthropic has launched OSS Scanner, a free, opt-in vulnerability scanner for critical open-source projects. The Anthropic OSS Scanner runs the company’s strongest models, including Claude Mythos, against enrolled codebases. Reports go straight to maintainers without human review. That trade delivers faster findings, but shifts triage work onto project teams. 11 days after the October 8 launch, Anthropic reported 77 critical and 803 high model-rated findings.
TL;DR
- Scale: 114 projects accepted from about 650 applications, as of October 19.
- Runs on: Anthropic’s infrastructure, free. Maintainers supply a Dockerfile; audits run in an isolated VM with no Internet.
- Performance: 85 of 97 audited critical and high findings (88%) met Anthropic’s disclosure bar.
- Best: Only 1 of 97 audited findings was a false positive. wolfSSL says 72 of 74 reports were valid.
- Worst: No human review before delivery. 11 of 97 audited findings duplicated known issues; maintainers report inflated severity.
What is Anthropic OSS Scanner?
OSS Scanner is a free application from Anthropic that periodically audits open-source repositories with its strongest models. The Frontier Red Team and Safeguards Research team run it together. The design borrows from Google’s OSS-Fuzz, but swaps fuzzers for language models.
The main goal is a review bottleneck. Over 6 months, Anthropic’s models surfaced over 29,000 candidate vulnerabilities. Human reviewers triaged roughly 6,000 of them. Nearly 5,000 more reports went directly to maintainers who asked for everything, validated or not. OSS Scanner turns that ad hoc request into an opt-in program.
How does OSS Scanner work?
Core maintainers enroll by opening a pull request on anthropics/oss-scanner. The PR adds projects/, based on the project template.
- Required fields:
repo,primary_contact, and a Dockerfile path. - Optional fields:
threat_model,pgpfor encrypted reports,auto_ccs,homepage, anddisabled.
The Dockerfile builds the project with network access. The VM then moves to a network without Internet for the audit. Agents in the pipeline double-check bugs, run root cause analysis and propose patches.
Each report contains a self-contained reproducer and an explanation of the bug. Where possible, it includes a bisection to the commit that introduced it. A candidate patch is attached when available.
An optional threat_model.md defines adversarial inputs, out-of-scope code and a severity rubric. Without it, the scanner guesses.
How accurate are OSS Scanner reports?
Anthropic team asked its penetration testers to check 97 critical and high findings across 48 projects.
- 85 (88%) met the bar for Anthropic’s coordinated vulnerability disclosure (CVD) process.
- 11 were real but duplicated known issues or other scan findings.
- 1 was invalid.
Maintainer data points in the same direction. wolfSSL received 74 reports; all but 2 were valid, and 5 became CVEs. curl patched CVE-2026-92392, a High-severity HTTP/2 upload use-after-free. It affects curl 8.12.0 to 8.22.0 and is fixed in 8.23.0, released October 14, 2026.
Anthropic links the timing to benchmark progress. On CyberGym, LLMs went from finding under 20% of vulnerabilities to over 85%. Its broader CVD dashboard lists 6,157 disclosed vulnerabilities across 591 projects, as of October 2. 516 of those are patched.
