News
What Is Anthropic OSS Scanner? Free AI Bug Scans for Open Source
Anthropic OSS Scanner is a free, opt-in service that uses Claude Mythos to scan open-source projects for security bugs. Who qualifies, how to enroll, the catch.
Anthropic OSS Scanner is a free service that points Anthropic’s strongest AI models at open-source code and emails the maintainers any security bugs it finds. It opened on October 8, 2026, and it is back in the headlines this weekend as more projects sign up.
If you saw "OSS Scanner" and wondered whether it is an app you can download, it is not. It is a program for the people who maintain important open-source software. Here is what it does, who can use it, and the one big catch.
What is Anthropic OSS Scanner?
OSS stands for open-source software: code anyone can read and reuse, like the libraries that run web servers, phones and routers. Anthropic OSS Scanner (also written "Anthropic OSS-Scanner" or "Claude OSS Scanner") is a vulnerability-finding service for that code.
Anthropic says it is modeled on Google’s OSS-Fuzz, a long-running free service that tests open-source projects for crashes. The difference is the engine. OSS-Fuzz throws huge amounts of random input at code to see what breaks. OSS Scanner uses language models that read the code, reason about how it could be attacked, and try to prove the bug is real.
It grew out of Project Glasswing, Anthropic’s earlier work using Claude to hunt for bugs. Anthropic says that over the last six months it found more than 29,000 candidate vulnerabilities in important software, manually reviewed about 6,000, and sent nearly 5,000 unverified reports straight to maintainers who asked for them. OSS Scanner turns that fast lane into a program anyone eligible can join.
Anthropic OSS Scanner at a glance
| Question | Answer |
|---|---|
| Made by | Anthropic, the company behind Claude |
| Launched | October 8, 2026 |
| Price | Free for enrolled projects |
| Models | Anthropic’s strongest models, including Claude Mythos |
| Who can join | Core maintainers of established, security-critical open-source projects |
| How to join | Pull request to github.com/anthropics/oss-scanner with a project config and a Dockerfile |
| What you get | Emailed reports with a reproducer, an explanation, and a candidate patch when available |
| Human review | None before the report is sent |
Is Anthropic OSS Scanner free?
Yes. Anthropic says enrolled projects get "thorough, periodic security scans" from its strongest models at no cost. There is no paid tier for it. Separately, Anthropic’s Claude for OSS program offers free Claude Max 20x plans to maintainers for fixing work, and Anthropic sells a different product, Claude Security, to companies.
Who qualifies?
Not every GitHub repo gets in. Anthropic reviews each request by hand and uses criteria similar to OSS-Fuzz. Its FAQ lists things like the project’s age, how often it is updated, its OpenSSF criticality score, an open-source license, stars and forks, how widely it is used, and how security-critical it is. Brand-new projects and most forks usually do not qualify.
Only core maintainers can sign up. Anthropic checks the security contact against sources like the project’s SECURITY.md file, and it re-checks eligibility before each report goes out, so reports only reach the people who own the code.
How to enroll in Anthropic OSS Scanner
- Fork the enrollment repo. It lives at github.com/anthropics/oss-scanner.
- Add a config file under projects/<your-project>/ with your git repository link and a primary contact email.
- Point to a Dockerfile that installs dependencies and builds the project. The build can use the internet, but the scan itself runs fully offline in a locked-down sandbox, so make sure your tests pass inside the container.
- Add optional extras: CC addresses, a GPG key to encrypt report emails, or a threat_model.md file that tells the agent what matters most.
- Open the pull request and wait for a human at Anthropic to review it.
To pause reports later, set disabled: true in the config. To leave, delete your project folder. The Hacker News counted 116 enrollment pull requests the day after launch.
The catch: no human checks the reports
This is the trade-off Anthropic is upfront about. OSS Scanner reports are fully model-generated, "without human review or triage," so some may be wrong, including the severity rating. Maintainers need the time to verify what lands in their inbox.
How often is it right? In an early test, expert penetration testers checked 97 critical and high-severity findings across 48 projects. Anthropic says 85 of them (88%) met its bar for disclosure, 11 were real but duplicates of known issues, and 1 was a false positive. That is Anthropic’s own number on its own sample, not an independent audit.
Two other rules matter. Unverified findings do not come with the usual 90-day disclosure clock, because they might be false alarms. And projects that want only human-checked reports can skip the program and stay in Anthropic’s regular coordinated disclosure process.
OSS Scanner vs OSS-Fuzz vs OpenAI Patch the Planet
| Anthropic OSS Scanner | Google OSS-Fuzz | OpenAI Patch the Planet | |
|---|---|---|---|
| How it finds bugs | AI models read and probe code | Fuzzing (random inputs at scale) | AI-assisted research with partners |
| Who signs up | Maintainers opt in via pull request | Maintainers opt in via pull request | Run by OpenAI with security partners such as Trail of Bits |
| Cost to projects | Free | Free | Free |
| Human review before report | No | Automated crash reports | Coordinated with maintainers |
OpenAI’s program is the one behind the recent PHP2Shell fix, where an outside researcher used GPT-5.6 Sol to find a critical PHP flaw. The two labs are now racing to show their models help defenders first.
What this means for you
For most people, OSS Scanner works quietly in the background. The open-source code inside your apps, browser and router may get checked by an AI agent, and bugs get fixed before attackers find them. Anthropic’s stated bet is that "in two years, AI will favor defense."
It is also a clear example of agentic AI at work: not a chatbot answering questions, but an agent that builds a project, explores it on its own, and hands back a proof and a patch. Anthropic also published the sandbox tool it uses to fence agents in, which we cover in What is Anthropic Sandbox Runtime?
You do not need to run a security lab to get some of that help. With Agentic AI on iPhone, you can paste in a security advisory and ask one agent to explain it, or put up to 4 agents on it so one writes the steps and another double-checks them.
Quick answers
What is Anthropic OSS Scanner? A free, opt-in program from Anthropic that uses AI models, including Claude Mythos, to scan eligible open-source projects for security bugs and email the maintainers.
When did Anthropic launch OSS Scanner? October 8, 2026.
Is Anthropic OSS Scanner free? Yes. Enrolled projects get periodic scans at no cost.
Can I scan my own app with OSS Scanner? Only if it is an established open-source project you maintain and Anthropic accepts it. Private code and brand-new repos are not what it is for.
Is it the same as OSS Document Scanner? No. OSS Document Scanner is an unrelated open-source Android app for scanning paper documents.
Are OSS Scanner reports accurate? Mostly, by Anthropic’s own test: 85 of 97 serious findings met its disclosure bar. But no human checks reports before they are sent, so maintainers must verify them.
Sources: Anthropic: Launching an opt-in vulnerability-finding service for open source · Anthropic: OSS Scanner FAQ · GitHub: anthropics/oss-scanner · The Hacker News · BetaNews · Google OSS-Fuzz
