Google pauses open-source bug bounty program over AI spam surge

By Splitfeed EditorialLast updated Oct 5, 2026, 13:00 UTC

In early October 2026, Google paused its Open Source Vulnerability Reward Program, known as the OSS VRP, halting product flaw submissions until 2027. The program, which pays researchers to find and disclose security vulnerabilities in Google-maintained open-source projects, was frozen after what Google described to TechCrunch as a "significant rise" in AI-generated reports. The submissions were characterized as invalid, hallucinated or otherwise low-quality, creating a backlog that left maintainers unable to process legitimate findings. Supply chain issue reports were carved out of the freeze and remain open, and Google is also reported to be tightening its submission rules alongside the temporary halt.

The freeze matters because bug bounty programs are a cornerstone of how major technology companies source independent security research. When the intake process becomes clogged, real vulnerabilities can sit unreviewed, which carries direct security implications for any software that depends on those open-source projects. The core dispute here is whether this outcome is a symptom of AI access being too broad, a structural weakness in how bounty programs handle volume, or simply an acceleration of a problem that already existed before generative AI tools became widely available.

The anti-AI camp has been the most vocal. Recep Cinet, posting on Bluesky, framed the dynamic precisely: reports are now cheap to produce but still costly to review, so the human triage queue becomes the bottleneck. The account ORZ summarized the broader frustration as "We're officially drowning in slop," and one commenter using the handle elmer2 argued that open access to AI has produced exactly this kind of outcome, suggesting it should be treated as a privilege rather than a commodity. The pro-AI camp had not published reactions to this story at the time of writing; typically, that camp would be expected to argue that AI tools also assist defenders and that bad actors represent a small fraction of use cases, not a reason to restrict access. The middle ground position, offered by the account Fennec Fox Fanfare, adds useful historical context: automated tools were already being misused to flood vulnerability databases like the CVE system before AI-assisted security testing arrived, making the current surge a continuation of an older problem rather than a fundamentally new one.

The clearest thing to watch is whether Google lifts the product flaw submission freeze on the schedule it has indicated, and what new submission rules accompany any reopening. If the tightened criteria successfully filter low-quality AI-generated reports while preserving access for legitimate researchers, that outcome would support the middle ground view that this is a solvable process problem. If the flood resumes or other major bug bounty programs announce similar pauses, that would give the anti-AI camp's structural argument considerably more weight.