Executive Summary

The meeting centered on refining security triage protocols, establishing new Artificial Intelligence (AI) access policies, and addressing infrastructure stability. The team aligned on several key decisions to improve project security and reduce the operational burden on maintainers.

A primary focus was addressing the high volume of AI-generated security reports. To mitigate this, the committee voted to adopt a stricter set of requirements for vulnerability submissions, mandating executive summaries and proven end-to-end exploits. Furthermore, the ESC agreed to transition AI access to a default-deny model and adopted Google’s centralized open-source guidelines for AI-generated code contributions, emphasizing human review and attribution.

The committee also reviewed several specific security vulnerabilities, categorizing them into actionable code fixes, configuration adjustments, or invalid reports. In addition to security matters, the committee prioritized addressing persistent Zuul CI resource flakiness and began planning for the upcoming deprecation of the Soy template system.

Gerrit Engineering Steering Committee (ESC) Meeting Summary

AI Review Permission Logic: Rollout Strategy at Google

The project is transitioning AI access from a “default allow” model to a “default deny” configuration. This shift requires a careful rollout strategy to avoid disrupting ongoing operations or creating gaps in coverage for Google and other users.

The initial implementation relied on Gerrit’s default-deny model, which broke existing workflows at Google. A seamless implementation is required to ensure that teams can explicitly opt-in without losing their necessary integrations during the transition period.

The AI Review permission was merged in February 2026 with Change 553501. It is now possible to grant the ALLOW permission to refs/* on All-Projects. While this is a no-op for now (as the default is ALLOW anyway), this step will facilitate a smooth migration to the default-deny mode.

Actions

[HJ] will engage with Google engineers to implement a project-level opt-in AI Review permission to All-Projects prior to the final implementation flip.

AI-Generated Code Contributor Policy

The rise in AI-generated code necessitates a formal policy for contributors and reviewers to ensure transparency, accountability, and proper guardrails within the Gerrit project.

It is difficult to technically enforce the identification of AI agents generating code, or to prevent them from submitting it under a contributor’s identity. However, human accountability remains a core requirement to maintain code quality and project integrity. This accountability is legally bound by the Contributor License Agreement (CLA) signed prior to gaining push access to gerrit-review.googlesource.com.

Actions

[HJ] noted that Google has defined and published a centralized open-source policy and guidelines for AI contributions. [HJ] will create a documentation change to incorporate these rules into the gerrit-review.googlesource.com contribution policy.

The policy will mandate human review, verification, and total responsibility and ownership of any contribution to the human user, even if part or all of the code was generated using AI.

Security Reporting and Triage Process

[LM] raised the issue of a significant surge in security notifications. These reports are largely repetitive “red herrings” generated by AI, lacking substance or unique findings.

This high volume creates a severe operational burden for the members of the ESC. The ESC members could use AI for automatic triaging and selection; however, allowing potentially dangerous security issue details to leak to an external LLM could represent a violation of the security embargo required until the issue is fixed.

Actions

[LM] will propose a change to the Security Issues Notification Process to define stricter rules for submissions.

Future reports must include all of the following information:
1. An executive summary of the issue from an E2E perspective.
2. A verified end-to-end user journey demonstrating how the issue can be exploited.
3. A clear risk assessment highlighting the type of vulnerability involved and its severity.
Reports missing these elements will be rejected as incomplete; reports that fail to demonstrate an actual security risk will be redirected to regular issues.
[HJ] will discuss internal AI triage processes with stakeholders.

The ESC agreed that triaging security issues should not be the sole responsibility of its members, but rather shared with the entire community of core maintainers.

Specific Security Vulnerability Reviews

The committee reviewed a backlog of specific security vulnerability reports. (Details cannot be disclosed in these meeting minutes for confidentiality).

Actions

P1 and P2 code issues will be handled in the dedicated security fixes repository, while minor issues will be addressed as private changes.

Infrastructure - Flaky CI Resources

The Gerrit project utilizes two CI environments:

gerrit-ci.gerritforge.com (Hosted by GerritForge and self-managed by the community)
ci.gerritcodereview.com (Hosted on Google Cloud and self-managed by the community)

The ci.gerritcodereview.com environment is currently suffering from resource shortages, which has resulted in persistent flakiness and disrupted development workflows.

Actions

[HJ] will take ownership of the action previously handled by Emily and will investigate the progress of the resource shortage issue, providing a status update before the next ESC meeting.

Soy Template Replacement

Google is moving away from the Soy template system, which Gerrit currently relies on for various email and landing page templates, as well as plugins like ITS-Base and the associated other ITS-* extensions.

The Soy template system was initially created as an open-source project; however, it will become closed-source and its maintenance will cease.

Actions

[HJ] will provide an update regarding the replacement system for Soy templates by early next week. A replacement template system needs to be identified and integrated to maintain current functionality. Google is managing the timeline for this deprecation.