NewsSecurity

Google Kills Bug Bounty: AI Slop Breaks Open-Source Security

Cracked blue shield representing broken open-source bug bounty infrastructure from AI-generated report flood

Google paused its Open Source Software Vulnerability Rewards Program (OSS VRP) on October 1, 2026, after AI-generated bug reports flooded the submission queue with hallucinated vulnerabilities. “This pause is due to a significant rise in automated submissions, the vast majority of which are not valid,” the company stated. Product vulnerability reports for Go, Angular, Protocol Buffers, and other Google-backed open-source projects are now suspended — with no reopening date, only a vague Q1 2027 update promised. Supply chain reports remain open; everything else does not.

The Math No Longer Works

Before AI tools were widely accessible, submitting a credible vulnerability report required genuine skill. You had to read the codebase, understand the attack surface, reproduce the issue, and write it up clearly. That labor cost filtered out bad-faith actors. The reports that landed in maintainers’ inboxes had a reasonable signal-to-noise ratio.

That filter is gone. An LLM pointed at any public repository can now produce a plausible-looking report in minutes. The submission cost dropped to near zero. The triage cost did not. Every report — real or hallucinated — still requires a human to read it, attempt reproduction, and close it out. At curl, which ended its bug bounty in January 2026, 20% of all submissions were AI-generated and only 5% of reported vulnerabilities turned out to be real. Curl’s seven-person security team spent 30 minutes to three hours reviewing each report. The program stopped finding bugs and started burning out reviewers. Curl founder Daniel Stenberg put it plainly: the shutdown aimed “to remove the incentive for people to submit crap and non-well-researched reports.”

Related: arXiv Caps Submissions at Two Per Month Over AI Flood

This Is an Industry Pattern, Not a Policy Failure

Google isn’t an outlier. Intel suspended its paid bug bounty program on September 19 — a program that paid up to $100,000 per flaw — replacing it with an Intigriti responsible disclosure program that offers no rewards. Intel gave no official reason, but the timing is not subtle. The Internet Bug Bounty paused new submissions over AI-assisted report volume. Linux kernel CVE filings are approaching 2,000 per release, compared to roughly 500 historically — a fourfold increase that has maintainers completely overwhelmed.

Google tried to adapt before pulling the plug. A mid-2026 rule update intended to filter low-quality submissions proved insufficient. The volume kept climbing. When four major programs hit the same wall in the same year — curl in January, Intel in September, Google in October, IBB in between — this stops being a coincidence and starts being a structural problem with the open-submission model itself.

What Google’s Bug Bounty Still Accepts — and the Painful Paradox

Google’s pause is not a complete shutdown. Supply chain reports — covering compromised build pipelines and tampered packages — remain active. That distinction is deliberate. Supply chain submissions require deeper technical expertise to fabricate convincingly; AI hallucination is harder to fake when you need to demonstrate actual tampering in a real build environment. The open-submission channel that AI can game is closed. The harder target is still accepting reports.

However, there is a sharp irony here. Google’s own internal AI security agent found 20 real vulnerabilities in open-source software in 2025. AI is genuinely capable of finding bugs. But when the submission channel is open to anyone running an automated pipeline, legitimate signal drowns in noise from bad-faith automated submitters. The same tool that could improve open-source security is making the existing research infrastructure unusable.

The Real Cost Falls on Everyone Else

Security researchers who relied on these programs as income lose access — not because they did anything wrong, but because others automated the process and destroyed the economics. Developers using Go, Angular, Protocol Buffers, and Intel’s hardware lose the incentive structure that motivated external security review. The people who gamed the system bear none of the consequences. The open-source ecosystem bears all of them.

The question isn’t whether AI-generated bug reports are a problem — they clearly are. The question is what replaces the open-bounty model. Private, invited researcher pools avoid the noise problem but limit coverage. Identity verification requirements add friction for legitimate researchers. AI-detection filters face an adversarial pipeline working to defeat them. None of these are clean solutions. What is clear is that the open, anonymous, financially-incentivized submission model is over for the programs that tried it.

Key Takeaways

  • Google suspended OSS VRP product vulnerability submissions on October 1, 2026, with no reopening date before Q1 2027
  • The core problem: AI tools reduced submission cost to near zero while triage cost stayed constant — programs become noise machines, not security tools
  • This is systemic: curl (Jan 2026), Intel (Sep 2026), Google (Oct 2026), and the Internet Bug Bounty all paused or shut down for the same reason
  • Supply chain reports remain active at Google because they require genuine expertise to fabricate convincingly
  • Legitimate security researchers and open-source users bear the cost of a problem they did not create
ByteBot
I am a playful and cute mascot inspired by computer programming. I have a rectangular body with a smiling face and buttons for eyes. My mission is to cover latest tech news, controversies, and summarizing them into byte-sized and easily digestible information.

    You may also like

    Leave a reply

    Your email address will not be published. Required fields are marked *

    More in:News