Google Tightens Open-Source Bug Bounty Rules After Surge in AI-Generated Vulnerability Reports

Google has stopped accepting new “product vulnerability” reports through its Open Source Software Vulnerability Reward Program (OSS VRP) due to a significant increase in automated submissions that are predominantly invalid.

This restriction took effect on October 1, 2026, and Google plans to provide an update regarding this part of the program in the first quarter of 2027.

The decision addresses a growing challenge for bug-bounty programs: AI-assisted or fully automated reports that may appear technically plausible but lack a reproducible security impact.

Google said the surge in submissions has forced maintainers and security engineers to spend considerable time validating reports that often turn out to be hallucinated, non-exploitable, or otherwise out of scope.

What Google Paused

The pause specifically affects new OSS VRP reports categorized as product vulnerabilities.

Previously, these included design or implementation flaws in Google’s open-source projects that could substantially compromise the confidentiality or integrity of user data in software builds using that code. Examples of past reports included:

• Memory corruption flaws in file parsers or network protocol implementations
• Path traversal vulnerabilities
• Failures in sanitization functions, including HTML sanitizers
• Insecure defaults or misleading code examples in project documentation

Reports submitted before October 1 remain unaffected and will continue to go through Google’s existing triage process. This decision does not impact OSS VRP reporting for software supply-chain compromises, which can involve the compromise of source code, build systems, release artifacts, package-manager accounts, or signing keys.

Google is directing researchers to submit eligible findings through other vulnerability reward programs instead of the paused OSS VRP product vulnerability channel.

Here are the recommended routes for different types of findings:

Finding type	Recommended route
Vulnerability affecting a Google Cloud product	Google Cloud VRP, for applicable repositories
Vulnerability involving an AI product or service	Google AI VRP
Security improvement or hardening contribution	Patch Rewards Program
OSS supply-chain compromise	OSS VRP supply-chain reporting remains available

Some Google Cloud repositories that affect Cloud products may still accept product vulnerability reports through the Cloud VRP. Therefore, researchers are encouraged to demonstrate a direct product impact and choose the reporting program that aligns with that impact.

This change does not signify a complete halt to Google’s open-source security efforts. Instead, it eliminates the highest volume of submissions while Google “reformats” the program. However, it raises the importance of evidence-backed reporting.

Researchers should include a buildable proof of concept against a current version, detailed reproduction instructions, affected versions, realistic attack scenarios, crash evidence where applicable, and a clear explanation of confidentiality, integrity, or supply chain impact.

For memory-safety issues, reproducible OSS-Fuzz results and a patch may be particularly valuable under Google’s established triage expectations.

The core security concern is that generative AI can lower the cost of identifying suspicious code patterns but can also produce numerous false positives.

A path that appears to be attacker-controlled in static analysis might be sanitized upstream, remain inaccessible to untrusted users, or require unrealistic deployment conditions. At scale, these mistakes create a triage burden that can delay the remediation of genuine flaws.

Google’s pause warns researchers and organizations using AI for vulnerability discovery. While AI can aid in code review, variant analysis, proof-of-concept development, and fuzzing workflows, it does not replace the need for validation. High-quality reports must demonstrate exploitability and impact, not merely highlight code that appears risky.

Until Google publishes its Q1 2027 update, researchers targeting its open-source ecosystem should focus on validated supply-chain findings, product-specific Cloud or AI impacts, and patch-based security improvements.

Stops Cyber threats before impact with 21 min faster MTTR. Integrate ANYRUN’s Sandbox in your SOC.