Google OSS VRP Product Bug Pause: Where to Send Your Finding Now

2026-10-04
Brian

· Updated
Subscribe
Get New Posts by Email
New RepoSquare posts in your inbox: comparisons, setup guides and explainers on open-source tools and AI coding agents. No spam, and you can unsubscribe anytime.

Explainer: this isn’t a bounty I’ve collected. It’s a reading of Google’s own OSS VRP rules and the public repository tier list, both of which I pulled from Google’s bughunters repository on 18 November 2026 and checked line by line.

If your scanner flagged a memory bug in a Google library last night, don’t file it to the OSS VRP. Since 1 October 2026, Google’s Open Source Software Vulnerability Reward Program no longer accepts product vulnerabilities. Supply-chain reports still pay, reports filed before that date are unaffected, and Google has promised an update in Q1 2027. So the real question now is which door your finding goes through, and the answer depends on what the bug lets an attacker change.

What Exactly Did Google Pause?

Google paused product vulnerabilities in its open-source code, and nothing else. The live rules page says Google is “no longer accepting product vulnerabilities submitted to the OSS VRP”, that it will “continue to reformat and work on this aspect”, and that some Google Cloud repos may still take product reports through the Cloud VRP. Google’s @GoogleVRP account gave the reason as “a significant rise in automated submissions, the vast majority of which are not valid.”

The motive is interesting, but it doesn’t change where you file. What changes things is the rules table, and that table splits findings into lanes.

LaneStatus after 1 Oct 2026
Product bugs in Google OSSNot accepted into OSS VRP
Supply-chain compromisesIn scope and paid
Other security issues, like credential leaksIn scope: $1,000 at OT0, $500 at OT1, nothing at OT2 or OT3
Reports filed before 1 Oct 2026Unaffected
Product bugs in some Cloud-tied reposMay go through Cloud VRP
Decision tree for a Google open-source bug: supply chain goes to OSS VRP, product bugs are paused with some Cloud-tied repos going to Cloud VRP, AI issues go to AI VRP, hardening patches go to Patch Rewards

Is My Bug a Product Bug or a Supply-Chain Bug?

Ask what the attacker gets. If the win is “I changed what users download” or “I can commit to main without review”, you’re in supply chain. If the win is “an app that links this library crashes or leaks”, it’s a product bug, and that lane is closed for now.

Google’s own supply-chain examples are concrete: pushing code to main without the usual approval gate, an insecure GitHub Actions or GCP build setup that lets you tamper with released artifacts, leaked package-manager publish credentials, and stolen signing keys. Where a “PR must be approved” gate exists, you have to show a way around it. Anything that only works after a maintainer approves your pull request counts as insider risk and earns credit at best.

Report sketchLaneWhere to file
Use-after-free in a Google parser library used by ChromeProductNot OSS VRP after 1 Oct
A GitHub Action in a Google repo can push to npm with a token printed in logsSupply chainOSS VRP
Prototype pollution in a demo under google-researchProduct, OT3Unpaid even before the pause
A misconfigured org webhook lets you add a deploy key on an OT1 repoSupply chainOSS VRP
Prompt injection in a Google AI productAI productAI VRP

Which Google Repos Still Count, and at Which Tier?

The tier list is public, and I checked it rather than guess. Google keeps the OT0 and OT1 repositories in a machine-readable file in its bughunters repo, along with a field that says where product bugs for each repo should go.

git clone --depth 1 https://github.com/google/bughunters.git
cd bughunters
grep -o "tier: [A-Z_0-9]*" oss-repository-tier/external_repositories.txtpb | sort | uniq -c
#     26 tier: TIER_OT0
#     47 tier: TIER_OT1
grep -o "product_vuln_scope: [A-Z_]*" oss-repository-tier/external_repositories.txtpb | sort | uniq -c
#      8 product_vuln_scope: SCOPE_CLOUD_VRP
#     65 product_vuln_scope: SCOPE_OSS_VRP

That second count was the useful surprise. I expected the “may still go through Cloud VRP” line in the rules to be vague. It isn’t. The file names exactly eight repos where product bugs are routed to the Cloud VRP: cdapio/cdap, cdapio/cdap-ui, the five google/adk repos for Python, JS, Go, Java and Kotlin, and google-gemini/gemini-cli. If your product bug is in one of those, you still have a paying door.

To check a single repo before you write anything, grep for it:

grep -A2 'url: "https://github.com/golang/go"' oss-repository-tier/external_repositories.txtpb
#   url: "https://github.com/golang/go"
#   tier: TIER_OT0
#   product_vuln_scope: SCOPE_OSS_VRP

If a repo isn’t in the file, it’s OT2 or OT3. OT2 has no published list and the panel decides, while OT3 covers samples, google-research and demos, and doesn’t pay at all. One limit to keep in mind: the file can change, and the last commit I saw touching the repo was on 10 November 2026, so pull it fresh before you rely on a tier.

How Much Does a Supply-Chain Report Pay Now?

Between $500 and $31,337, depending on the tier. The reward table in the rules source reads like this today.

CategoryOT0 FlagshipOT1 ImportantOT2 StandardOT3 Low
Supply chain$3,133.7 to $31,337$1,337 to $13,337$500 to $3,133.7none
Product bugsnonenonenonenone
Other security issues$1,000$500nonenone

The product row is dashes across every tier, which is the pause written as a price list. The panel can also add a bonus of about $1,000 for particularly clever findings or well-written reports.

Google OSS VRP supply-chain reward ranges by tier: OT0 $3,133.7 to $31,337, OT1 $1,337 to $13,337, OT2 $500 to $3,133.7, OT3 no amount, product cells all dashes

Where Should I Send a Finding This Week?

Match the finding to the program before you open any form.

FindingWhere
Supply chain on Google OSSOSS VRP, with the repo URL
Product bug in one of the eight Cloud-scoped reposCloud VRP
Google AI product issueAI VRP when in scope
Hardening patch with no bounty fitPatch Rewards Program
Bug in a third-party dependencyFix upstream first, wait at least 30 days after the upstream fix, then show it triggers in Google OSS

Don’t run noisy scanners against shared Google infrastructure, and test locally when you can. Bugs in platforms Google doesn’t own, like the code host or a public CI service, stay out of scope except for Google’s own configuration of them.

What Does a Report That Gets Paid Look Like?

Short, reproducible and specific about what an attacker could alter. I keep this checklist in my notes before opening the form, because empty lines are exactly what gets bulk filings closed.

OSS VRP report checklist (supply-chain lane)

1. Lane:   [ ] supply chain   [ ] NOT a paused product filing
2. Target: repo URL, tier from the tier file, commit or package version
3. Impact: "An attacker without maintainer rights can ..."
4. Preconditions: public fork? stolen token? which approval gate is bypassed, and how?
5. Proof: exact steps on a recent build, workflow YAML or commands,
          which artifact or branch could be altered, what you did NOT do
6. Fix sketch: required reviewers, OIDC, key rotation, config change
7. Timeline: discovered / upstream fixed + 30 days waited / reported

A subject line like “GitHub Actions in org/X can push to PyPI without environment protection” tells a reviewer the lane and the impact in one read. “AI found SQL injection in example/” tells them neither, and treating “AI found it” as a severity bonus usually works the other way.

Why Did Bounty Programs Get Stricter?

Because maintainers are drowning in low-signal automated reports, and Google’s pause is one response to that. Other open-source maintainers, including curl’s Daniel Stenberg, have written publicly about bounty inboxes filling with AI-generated noise. Even before October, OT0 and OT1 memory bugs needed exact OSS-Fuzz reproduction steps or an already merged patch.

If you’re a maintainer, the same discipline helps you read your own inbox, and How to Review AI-Generated Code Faster covers the reading side. If your finding is in an agent tool rather than a library, How Do I Check If a Claude Code Mod Is Safe Before Installing? shows the kind of local proof reviewers trust. For how labs separate research demos from production targets, see Gemini 4 Argon Anthropic’s AI worm measurements and Mistral Large 4, and for why agent stories need a measured scope, Claude’s unintended actions.

If you already filed a product bug before 1 October, leave it alone and reply in that thread if Google asks for more. If you’re holding one now, check the tier file for those eight Cloud repos, and if it isn’t there, put it in your notes until Q1 2027 rather than burning trust on a closed lane.

Get New Posts by Email
New RepoSquare posts in your inbox: comparisons, setup guides and explainers on open-source tools and AI coding agents. No spam, and you can unsubscribe anytime.

1 comment

Leave a comment