Google has officially suspended its Open Source Software Vulnerability Rewards Program (OSS VRP), citing a massive increase in low-quality, automated submissions. The suspension, which went into effect on October 1, 2026, highlights a growing crisis in the cybersecurity industry: the rise of AI-generated "slop" that overwhelms human triage teams and maintainers.
The company announced the pause through its official Bug Hunters website and social media channels, explaining that the program will remain frozen until the first quarter of 2027. During this time, Google intends to re-evaluate the program's structure to better handle the changing landscape of security research in the age of generative artificial intelligence.
What happened?
Google's Open Source Software Vulnerability Rewards Program was designed to incentivize security researchers to find and report vulnerabilities in the company's vast array of open-source projects. These projects include critical infrastructure and developer tools such as Golang, Angular, and Bazel. By offering financial rewards for legitimate bug reports, Google aimed to harden the security of the broader internet ecosystem.
However, the program has become a victim of its own success, or more accurately, a victim of the ease with which users can now generate plausible-sounding security reports using large language models (LLMs). Google confirmed that the pause is a direct response to a "significant rise" in automated submissions.
According to statements from the company, the vast majority of these AI-generated reports are not valid. While they may appear professional or technically dense at first glance, they often describe non-existent vulnerabilities or contain "hallucinations," a term used when AI models confidently state incorrect information.
The burden on maintainers and engineers
The primary issue is not just the inaccuracy of the reports, but the sheer volume. In a typical bug bounty workflow, every submission must be triaged by a human engineer. This process involves reading the report, attempting to reproduce the bug, and determining if it poses a genuine security risk.
When thousands of automated, low-quality reports flood the system, the signal-to-noise ratio drops significantly. Google engineers and open-source maintainers have found themselves spending an increasing amount of time debunking "hallucinated" vulnerabilities rather than fixing actual security flaws.
This phenomenon, often referred to as "AI slop," poses a systemic risk to the cybersecurity community. If engineers are tied up reviewing fake reports, real vulnerabilities might go unnoticed for longer periods. For the open-source community, which often relies on a small number of overworked maintainers, this influx of automated noise can lead to burnout and a decrease in overall project security.
The problem of AI hallucinations in security
The rise of generative AI has made it easier than ever for individuals to participate in bug bounty programs. While this lower barrier to entry could theoretically lead to more eyes on code, the current reality is a surge in "beg-bounties," where participants use automated tools to spam programs in the hopes of a quick payout.
AI models are particularly adept at mimicking the structure and tone of a legitimate security report. They can use the correct terminology, reference specific lines of code, and suggest remediation steps. However, these models do not actually "understand" security logic; they are predicting the next likely word in a sequence. This results in reports that:
- Describe vulnerabilities that do not exist in the code.
- Misinterpret the severity of minor bugs.
- Suggest "fixes" that would actually break the software or introduce new security holes.
- Reuse old, patched vulnerabilities found in the model's training data.
Google's decision to freeze the program suggests that the current automated filtering systems are insufficient to distinguish between a sophisticated human researcher and a poorly prompted AI bot.
Broader industry trends
Google is not the first organization to struggle with the impact of AI on bug bounty programs. Throughout 2025 and 2026, cybersecurity experts have warned that the economics of bug bounties are being disrupted. Because it costs essentially nothing for a user to generate and send a hundred AI-written reports, the incentive to "spray and pray" is high, even if the success rate is near zero.
Major bug bounty platforms and large technology firms are currently in a technological arms race, attempting to develop AI-based triage systems that can detect and filter out AI-generated submissions. Ironically, the solution to the AI slop problem may involve using even more AI to act as a gatekeeper.
What happens next?
Google has stated that it will provide an update on the status of the OSS VRP in the first quarter of 2027. It is likely that the program will return with stricter submission guidelines, possibly including:
- Requirements for proof-of-concept code that can be automatically verified.
- Stricter penalties or bans for users who submit a high volume of invalid, automated reports.
- Integration of more advanced triage layers to filter submissions before they reach human engineers.
In the meantime, Google is encouraging legitimate security researchers to focus their efforts on its other active bug bounty programs. These programs, which cover products like Chrome and the Android ecosystem, have different submission criteria and triage processes that may be better equipped to handle the current influx of automation.
The pause serves as a clear signal to the security community: while AI can be a powerful tool for finding bugs, its current use in generating bulk submissions is creating more problems than it solves. The industry must now find a way to preserve the benefits of crowdsourced security while mitigating the noise of the AI era.
Filed under: AI, TechNews, Cybersecurity, Software, Google, OpenSource