Google’s Open Source Bug Bounty Program Has a Big AI Problem

Bug bounty programs are intended to be mutually beneficial: large organizations can plug holes in their security while hackers skilled enough to discover vulnerabilities are paid for their efforts. However, with the advent of AI, many such programs have been inundated with poor quality reporting generated by LLMs rather than human researchers, and often refer to fictitious, unexploitable, or hallucinated vulnerabilities.

Rather than helping, it’s a hindrance that recently caused Google to join Curl, Intel and HackerOne and suspend its paid bug bounty program, called Open Source Software Vulnerability Reporting Program – OSS VRP for short. These were the highest-paying bounty programs, meaning ethical hackers now have fewer ways to profit from their talents. Google previously awarded up to $31,337 for the most serious reported vulnerabilities, with Intel up to $150,000.

However, OSS VRP is not lost for good. Google says that even though the vast majority of automated submissions are invalid, it has committed to reformatting the program and providing an update in the first quarter of 2027. In the meantime, bounty hunters are encouraged to continue submitting reports to Google’s other VRP or Patch Rewards programs, which remain active. Alternatively, you can try to claim Apple’s $2 million bounty for discovering exploits that could achieve the same goals as a sophisticated mercenary spyware attack.

You may be surprised to learn that most of what you see online is now AI-generated, including articles, music, social media posts, and comments, all “written” by LLMs. Artists rightly criticize much of this avalanche, calling it mediocre at best, but it’s also a problem for AI companies. Researchers warn that AIs typically deteriorate when learning from AI-generated content, a phenomenon known as the “garbage in, garbage out” rule. Since AI tools like ChatGPT are trained on publicly available information across the internet, including on platforms like Reddit, this creates a potentially counterproductive cycle. This is one of the reasons Google and Anthropic have embarked on major projects to digitize physical books: they are pristine content, written entirely by humans, from a time before AI was commonly used.

This is not to say that there aren’t legitimate uses for AI. It has proven extremely effective in detecting certain forms of cancer, improving accessibility via live transcription and captioning, and creating new drugs for diseases like Parkinson’s. It even has a place in bug hunting, having recently identified over 1,000 vulnerabilities in Linux. In comparison, low-quality reports generated by people who are not cybersecurity professionals are often submitted without additional work to identify a true vulnerability resulting from an AI error.

AI tools are set to become even more ubiquitous, freely available, and capable of generating an infinite amount of content that is correct at first glance. Abuse is likely to spread as well, and humans have not yet caught up. Companies like Google will likely need to implement even more quality control systems and devote resources to searching for the few useful nuggets among the growing sea of ​​uselessness.