AI - Anthropic launches OSS Scanner, free vulnerability scans for critical open source projects
Anthropic has launched OSS Scanner, a free service that uses its strongest models to periodically scan critical open source projects for vulnerabilities. Maintainers get the findings by email, complete with a proof-of-concept and, where possible, a suggested patch. Free security help for overstretched maintainers is very welcome, although this is very much aimed at established projects rather than your weekend side project 🤣
OSS Scanner
Official announcement
We launched OSS Scanner, which gives critical open source projects free, periodic vulnerability scans from our strongest models.
Each finding includes a proof-of-concept, and a patch where possible.
OSS Scanner - Anthropic Frontier Red Team
From the official page:
OSS Scanner is a service by Anthropic to scan open-source repositories for security vulnerabilities. It’s an opt-in service informed by our experience using Claude to find vulnerabilities during Project Glasswing. Projects that join will receive thorough, periodic security scans by our strongest models at no cost.
What is new?
Anthropic already scans open source software and reports vulnerabilities after human review through its coordinated vulnerability disclosure (CVD) process. It says it had reviewed more than 6,000 such reports by October 2026. The catch is that human review takes time.
OSS Scanner is an optional fast track:
- Free, periodic scans of enrolled projects, with Anthropic covering the full cost.
- Reports straight from the models, sent by email as soon as a scan completes, without waiting for human review.
- Proof-of-concepts, root cause analysis and candidate patches, produced by a pipeline that includes agents to double-check bugs and propose fixes.
- Repeat scans to look for newly introduced issues and anything missed previously. Frequency depends on factors such as demand and how widely a project is used.
- Offline, sandboxed scanning: the project’s Dockerfile is built with network access, then the agents audit the code without internet access.
- A threat model file, so maintainers can explain what is in scope, how severity should be rated and what kind of patches or proof-of-concepts are useful.
Who can sign up?
This is not for every repository. Anthropic says it uses criteria similar to Google’s OSS-Fuzz, focusing on established projects with a critical impact on infrastructure and user security. It considers factors such as exposure to untrusted input, the number of users and dependent projects, the OpenSSF criticality score, project age, update frequency and licensing.
Common exclusions: While we’re excited by brand new projects that want to build secure foundations from day one, they do not typically qualify. We typically do not scan forks of projects unless they are distinctly unique from the original project and independently qualify.
Core maintainers enrol by opening a pull request against anthropics/oss-scanner that adds a projects/<project>/project.yaml file. The required fields are the repository URL, a primary contact email and a Dockerfile that builds the project and installs everything needed for offline analysis. Every request is reviewed by a human, who checks that it comes from a core maintainer and that reports will go to a genuine security contact.
These reports are unvalidated. OSS Scanner sends findings before a human has reviewed them, so false positives are possible. Anthropic says the service is designed for projects that already keep up with verified high and critical vulnerability reports.
Because the findings are unvalidated, Anthropic says it does not apply a 90-day disclosure deadline to them. If a report is later validated through its standard CVD programme, disclosure may follow 90 days after the maintainer is told it has been validated. Maintainers can pause reports by setting disabled: true in their config, or opt out completely by removing their project directory.
If you are a maintainer and want to apply, read the eligibility criteria, template and terms before opening a pull request.
My take
I really like this. Open source maintainers carry a ridiculous amount of the world’s software on their shoulders, often unpaid, and anything that gives them high-quality security help without a bill attached is a good thing 😊
The interesting design choice is the trade-off between speed and certainty. Anthropic is explicit that unvalidated reports may contain false positives, and it is not forcing a disclosure clock onto them. That feels like the right approach. An AI-generated report with a proof-of-concept and suggested patch could save maintainers a lot of work, but it could also add to the pile if the findings are noisy. The eligibility bar is clearly there to send this to projects that can actually handle the reports.
It is also worth remembering that a suggested patch is still a suggestion. Maintainers will need to review it like any other contribution 😉
Relevant X post
Ziwen sums up the excitement nicely, and the eligibility point too:
Claude just did all of us in open source a favor!!
OSS Scanner is free. Established projects apply, then Claude scans their code for security holes and emails the maintainer a suggested fix. Anthropic says brand new projects usually don’t get in.
I might going to opensource serein browser and try to apply this week.
Let’s see if that counts as established.