← All insights
ECOSYSTEMZcash Foundation

Zebra's AI-assisted contribution policy

AI tools have increased contributions, but Zebra's principle that every change gets strict human review is unchanged. It explains how the project handles the growing number of external PRs.

The Zcash Foundation described Zebra's approach to AI-assisted contributions. It explained that AI coding tools have let contributors unfamiliar with Rust or the Zcash protocol make meaningful contributions, and that Zebra made four major releases in the past three months: 3.0.0, 3.1.0, 4.0.0 and 4.1.0. The Foundation said it will embrace this change but manage it deliberately. Every change, whether written by a human or with AI help, goes through strict human review by the engineering team.

Zebra is a consensus node implementation that validates every transaction and enforces the rules that protect users' financial privacy, and it will become the network's main consensus implementation after zcashd is deprecated and NU7. A consensus validation bug can reject a valid shielded transaction or accept an invalid one. What has changed is the volume of external pull requests. Some contributions are excellent, but some PRs come with no context or prior coordination and it is unclear whether the author understands the change. With maintainers' review capacity finite, it explains, opening a PR has a real cost.

The Foundation introduced three things. CONTRIBUTING.md requires contributors to first describe the change in an issue and wait for the team's answer, to disclose which AI tools they used and how, and to be able to explain the logic and design trade-offs of every change if asked during review. It also states the standard that PRs with no prior discussion, that were not requested, or that the author cannot explain may be closed. It also adopted AGENTS.md for AI agents, which guides them on contribution gates, crate structure, code patterns such as Tower service limits and error handling, and security constraints such as bounded allocation and input validation. Copilot code review instructions were also added.

The Foundation judged AI-assisted contributions to have been a net benefit for Zebra, while stressing that the engineering team deeply reviewed all of these features against the protocol specification. AI speeds up writing, it says, but understanding and responsibility stay with humans. To people who want to contribute, it asks them to start a conversation through an issue or Discord, and says AI tools may be used but should be disclosed, with the contributor understanding the content and respecting the process. Features that do not fit a consensus node can be handled in the Z3 stack, such as Zaino, Zallet and librustzcash. The Foundation said it will watch the effect of the guidelines for a few weeks and adjust.

This summary is based on the original; check the original for exact details.