Google pauses open-source bug bounty submissions after surge in invalid AI-generated reports

Google has paused a major part of its open-source bug bounty program after a surge in automated submissions, most of which the company says were invalid.

The suspension took effect October 1 for product-vulnerability reports submitted through Google’s Open Source Software Vulnerability Rewards Program, or OSS VRP. Google said it would provide an update during the first quarter of 2027 and encouraged researchers to use other Google vulnerability-reward programs or submit security patches instead.

The move is not a finding that artificial intelligence cannot discover real vulnerabilities. It is a response to the cost of evaluating a rapidly expanding volume of reports that may look plausible but do not demonstrate a practical security problem.

Google’s own program guidance describes several recurring issues: AI-generated reports that hallucinate how a vulnerability could be triggered, and reports that identify a coding mistake without showing meaningful security impact or a reachable attack path. In both cases, the burden falls on human engineers and open-source maintainers to reproduce the issue and determine whether it matters in practice.

That distinction is important because a bug bounty report is not the same thing as a confirmed vulnerability. Google’s rules ask researchers to provide details such as an affected version, the security impact, an attack scenario and, where possible, a buildable proof of concept. Those requirements are designed to separate a theoretically suspicious line of code from a flaw that an attacker can actually exploit.

Google had already tightened the OSS VRP earlier in 2026. In a March update, the company said it had seen a large increase in AI-generated reports and began requiring stronger evidence for some submission tiers, including an OSS-Fuzz reproduction or a merged patch. Later rule changes removed monetary rewards or credit for certain lower-tier product-vulnerability and other-security-issue reports.

The program covers security issues in open-source software maintained in Google-owned public repositories and selected repositories hosted elsewhere. Its scope includes projects and components whose security can affect developers well beyond Google itself, making the quality of incoming reports relevant to the broader open-source supply chain.

For researchers, Google’s pause changes the incentive structure. A report that merely points to a possible defect is less useful under a system demanding reproducible evidence. Google’s separate Patch Rewards program instead pays for proactive security improvements that are typically submitted as merged changes to eligible open-source projects. Listed rewards range from hundreds of dollars for smaller improvements to as much as $15,000 for complicated, high-impact work, with some temporary multipliers available for qualifying memory-safety improvements.

The pause also highlights a larger problem for vulnerability-response programs: AI can reduce the cost of producing a report without reducing the human cost of validating it. If automated tools generate findings faster than security teams can reproduce and prioritize them, the result can be a backlog that makes it harder—not easier—to identify serious flaws.

Google’s next update will indicate whether the company reopens the affected intake process with stricter evidence requirements, adds more automated filtering, narrows the program’s scope or keeps some categories closed. Until then, the company’s guidance points researchers toward higher-quality, reproducible submissions and patch-based contributions rather than high-volume speculative reports.