Anthropic OSS Scanner: Free AI Bug Reports With No Human Review
Agentic AIAnthropic’s new open-source security service makes a deliberate trade: it removes the human from the loop to get vulnerability reports to maintainers faster, and it accepts that some of those reports will be wrong. That choice says more about where AI-driven security is stuck than any accuracy figure does.
What Happened
On October 8, Anthropic launched Cyber Mission, a long-term program with two initial focus areas: critical infrastructure and open-source software. The infrastructure half, the Critical Infrastructure Defense Program, pairs frontier Claude models, on-site engineers, and threat research with 11 founding partners, including CrowdStrike, Palo Alto Networks, Dragos, Rockwell Automation, and Deloitte. It starts with a small cohort of providers that secure operational technology such as power grids, water systems, and factories.
The open-source half is OSS Scanner, a free, opt-in service modeled on Google’s OSS-Fuzz. Enrolled projects get periodic scans from Anthropic’s most capable models. Each report includes a proof of concept showing how the bug could be exploited, an explanation, and a suggested fix where one is available. Core maintainers of critical projects enroll through the oss-scanner repository on GitHub. As The Decoder reported, the reports are sent without human review and may contain errors, such as wrong severity ratings.
Why It Matters
The bottleneck in AI vulnerability discovery is no longer finding bugs. It is verifying and fixing them. Anthropic says as much: exploiting vulnerabilities has become cheaper, while verification and remediation remain slow. In Project Glasswing, where partners scanned hundreds of widely used open-source projects and humans triaged the results, months often passed between discovery and fix.
OSS Scanner is Anthropic’s answer to that gap. Skipping human review gets findings to maintainers faster, and the expected true-positive rate above 90% is the bet that this speed is worth the noise. The eligibility rule shows where the risk lands: the service is intended for projects that can keep up with the findings. Everyone else keeps receiving slower, human-verified disclosures under Anthropic’s coordinated vulnerability disclosure policy.
Anthropic is also funding the maintainers who will absorb this work. It cites support for the Python Software Foundation, Alpha-Omega and OpenSSF through the Linux Foundation, and the Apache Software Foundation. It also offers free Claude Max subscriptions through Claude for Open Source.
The Pattern
This is the fourth Stack Archive post since June on AI and open-source security, after Project Glasswing’s results, the Akrites patch-rate effort, and the Mythos 5 near-miss on an open-source project. The recurring angle is the same each time: AI finds problems faster than volunteer maintainers can absorb them.
What This Means for Builders
For maintainers, enrolling is a capacity decision, not a security checkbox. A 90% true-positive rate still means roughly one report in ten could waste triage time, and every report demands a response. Projects with a single part-time maintainer may get more value from the human-verified path.
For teams that depend on open source, the program shifts some upstream patching earlier, but only for projects that opt in and keep up. Dependency inventories and scanners still matter, because coverage will be uneven by design.
The So What
Anthropic predicts AI will favor defense within two years, while admitting that may not hold in the near term. OSS Scanner shows the reason for the hedge: frontier models can already generate findings at a pace humans cannot review. The useful question for any team adopting AI security tooling is not how many vulnerabilities it finds, but who triages them, how fast, and what happens to the 10% that are wrong.
Source: Anthropic
Content created with AI assistance and reviewed for accuracy.
Join the conversation
Stack Insiders is our free community for readers who want to go deeper — share resources, ask questions, and connect with others across every vertical we cover.
Join Stack Insiders →