AI Slop Bug Bounty Pauses: Google Follows Curl and Turso

What happened

Google has paused part of its open-source bug bounty, and it blames AI-generated reports. On October 1, 2026, Google's Vulnerability Reward Program account said it was "temporarily no longer accepting OSS VRP product vulnerability submissions." The stated reason was "a significant rise in automated submissions, the vast majority of which are not valid." 25 The company has promised an update on the program's future in the first quarter of 2027. 25

The pause is narrower than much of the coverage suggested. Reports filed before October 1 will still be processed. Supply chain reports, which cover compromised build pipelines and tampered packages, are unaffected. Some repositories tied to Google Cloud products can still be reported through the Cloud VRP. 5 What has stopped is the route most researchers used: find a flaw in one of Google's public projects, write it up, and get paid. 5 Coverage summed it up as "slop's fault," and that framing is accurate in spirit even though it overstates how much of the program closed. 2

Google is following smaller projects

Google's decision drew the attention, but smaller projects reached the same point first.

Curl, the widely used data-transfer tool, ended its HackerOne bounty on January 31, 2026. 3 One account dates the shutdown to February, which is a small discrepancy about when it took effect. 4 Founder Daniel Stenberg reported that the share of submissions that turned out to be real vulnerabilities fell from above 15 percent to below 5 percent during 2025. In his words, "not even one in twenty" reports was genuine. 3 Over its lifetime, the program confirmed 87 vulnerabilities and paid out more than US$100,000. 3

Turso, a SQLite-compatible database written in Rust, more recently retired a program that paid $1,000 per critical vulnerability. It had run for about a year. 1 The maintainer's explanation, titled "The wonders of AI," described spending most bounty hours reading LLM-written reports that no person had really meant for him to read. 1 The same pressure appears outside bounty programs. Popular GitHub repositories are receiving drive-by pull requests with AI-generated typo fixes and invented refactors. Issue trackers are filling with vague "there might be a bug" reports that include no reproduction steps. 1

The problem is shifting from bad reports to too many reports

The sources differ on one important point. Several describe the problem as mostly junk. Google says most of what it received was invalid. 2 Curl's confirmed rate fell sharply. 3 One analysis notes that an LLM can produce a convincing report in minutes, complete with a severity rating and a proof of concept that usually fails. 5

Stenberg's more recent account complicates that picture. According to one interview, curl reopened its bounty about a month after closing it, because report quality had improved. Volume, however, kept rising. 4 The project now receives an AI-generated report roughly every 18 hours, compared with about one a week before AI tools. Many of these reports are technically accurate. 4 Stenberg's view is that AI is good at finding bugs but weak at judging how severe they are or writing correct fixes, so the hardest work still falls on maintainers. 4 Duplicates are a growing problem too, because different researchers give the same model similar prompts and get the same findings. 4

One security firm's triage guide makes a related point: AI findings are not automatically worthless. The useful test is whether a report comes with a reproducible proof, not how many reports arrive. 3

Why it matters

The main issue is cost. Generating a report has become almost free, but verifying one still takes a skilled person's time. A bounty program works only when payouts attract more real findings than the noise that comes with them costs to sort through. AI tools have upset that balance in two stages. First came a flood of hallucinated reports. Now there is a flood of plausible reports that still need careful human review.

Google can afford triage staff far more easily than a volunteer like Stenberg or a small database team, which makes its pause more telling. When the best-funded program cannot cope with the inflow, the volunteer-maintained projects that much of the software supply chain depends on are under even greater strain. Keeping supply chain reports open suggests Google is protecting the area where the security stakes are highest.

Outlook

These pauses look like a reset rather than the end of bug bounties. Programs will probably return with stricter rules: required reproduction steps, some cost or reputation requirement to submit, deduplication, and perhaps AI-assisted triage to counter AI-assisted reporting. Curl's quick reopening suggests that model can work. 4 The open question is whether maintainer time can keep up as AI tools make real findings, not just fake ones, cheaper to produce. Google's update in early 2027 will show which direction the largest program takes. 25