Anthropic says its systems have identified more than 29,000 candidate vulnerabilities in some of the world's most important software projects — and it is now offering a free, opt-in scanner to open-source maintainers to help turn those findings into fixes.
What OSS Scanner does and how Anthropic built it
Anthropic unveiled OSS Scanner as an "opt-in service informed by our experience using Claude to find vulnerabilities during Project Glasswing," the company said. The scanner uses the firm's strongest models, including Claude Mythos, to produce fully model-generated security reports without routine human review or triage, a design choice Anthropic says will permit faster and more frequent scans.
The company framed OSS Scanner as part of a wider Cyber Mission that also includes a newly announced Critical Infrastructure Defense Program. Anthropic described that program as aimed at safeguarding critical infrastructure and open-source software, arguing that as AI lowers the cost of finding and exploiting bugs, defenders need advanced tools to accelerate fixes and explore more secure architectures.
Enrollment, configuration and the offline audit model
Core maintainers enroll by opening a pull request on the OSS Scanner GitHub repository and adding a YAML configuration file. The YAML must include:
- a link to the git repository to be cloned;
- an email address for the primary contact; and
- a repository-relative path to a Dockerfile that sets up the environment, pre-installs all dependencies and builds the project so an offline agent can conduct its security audit.
Anthropic emphasized that "the Dockerfile configures the environment that the project will run in and installs all dependencies so that the agent can perform its security audit without any internet access," and recommended maintainers verify that test cases pass inside the built container. Optional YAML fields include additional CC email addresses, a project home page, a GPG public key for encrypted report emails, a repository-relative path to a threat model file named "threat_model.md," and a "disabled: true" flag to opt out of receiving bug reports.
Your scanner finds 4,000 vulns. Which 12 matter?
Nubivance is a Rapid7 Registered Partner delivering vulnerability management as a service - scanning, risk-based prioritization, and remediation follow-through across IT and OT.
Fix the backlogFindings to date and Anthropic's disclosure posture
Anthropic reported that its work has identified more than 29,000 candidate vulnerabilities and that it has reported a little more than 6,000 flaws to maintainers. Those reports have yielded 584 advisories as of October 2, 2026. As of this writing, 116 pull requests to enroll projects in OSS Scanner have been submitted.
Unlike many cooperative vulnerability disclosure programs, Anthropic said it does not intend to impose a blanket 90-day disclosure period on findings from OSS Scanner because of concerns about false positives. The company added: "If we later validate one of these reports manually through our existing CVD program, we may disclose it under our CVD policy starting 90 days from when you are notified that a human has validated this report." Anthropic also left open the possibility that "as we gain greater confidence in OSS Scanner's performance, we may in the future impose a disclosure period on some high-severity vulnerability reports."
How the approach differs from automated bug-finding and from OSS-Fuzz
Anthropic said it expects to use a set of criteria similar to Google's OSS-Fuzz when selecting projects, while allowing the process to evolve. The company highlighted two key distinctions: OSS Scanner's outputs will be "fully model-generated" for speed and cadence, and the scanner accepts a build-time Dockerfile so the agent can perform audits without internet access. Together, those choices favor repeatable, containerized scans over live online fuzzing that reaches outward for dependencies or networked services.
What this means for open-source maintainers, defenders, and bad actors
Open-source maintainers: Projects that opt in must provide a Dockerfile and contact details; maintainers will receive direct, model‑generated reports and can decide how to triage them. Anthropic allows a "disabled: true" opt-out flag, and recommends validating test cases inside the submitted container.
Defenders and security teams: The Critical Infrastructure Defense Program and OSS Scanner aim to put model-powered detection tools into defenders' hands, with Anthropic forecasting that "in two years, AI will favor defense: it will be easier to catch bugs before they ship, write fundamentally secure software from scratch, and actively defend systems with models."
Bad actors: Anthropic explicitly cited the concern that "AI increasingly equipping bad actors to discover and exploit vulnerabilities" is a driver for the initiative, and framed OSS Scanner as a defensive countermeasure to accelerate fixes and reduce the window of exposure.
Anthropic's OSS Scanner ties an ambitious model-first approach to a familiar open-source enrollment flow: a pull request and a Dockerfile. The program already reports thousands of candidate flaws and hundreds of advisories, but the company has signaled it will keep adjusting disclosure rules and selection criteria as the scanner's accuracy and confidence improve. For maintainers weighing participation, the near-term tradeoff is between receiving fast, model-generated reports and managing the potential noise of automated findings without a guaranteed human triage step.
Source: https://thehackernews.com/2026/10/anthropic-launches-free-ai.html
