AI usage must be explicitly disclosed.
Contributors must clearly state:
- which AI tool(s) were used (for example: Claude Code, Cursor, Copilot)
- the extent to which the contribution was AI-assisted
Undisclosed AI usage in pull requests will result in closure.
If maintainers reasonably suspect undisclosed AI use, the pull request will be closed. Repeated incidents can result in a ban.
The author of the PR must write the description in part to demonstrate their understanding of the chagne.
All AI-assisted code must be fully verified by a human contributor.
Contributors are responsible for ensuring:
- the code has been run and tested
- the code behaves correctly in practice, not just in theory
It is recommended to attach test artifacts, such as screenshots or recordings, that demonstrate your code working for each test step.
AI assistance is permitted in issues and discussions, with a strict human-in-the-loop requirement:
- AI-generated content must be reviewed and edited by a human before submission.
- Verbosity, noise, and speculative content must be removed.
- Contributors remain responsible for research, accuracy, and clarity.
Contributors retain responsibility.
AI may assist with explanations, summaries, or drafting, but must not replace understanding or judgement. Having an AI post for you will result in a request to stop, repeat incidents will result in a ban.
Contributors should endeavour to understand and test any claims sourced from AI, when feasible. Untested claims must be flagged as such.
AI-generated media is not permitted.
This includes, but is not limited to: art, images, videos, and audio.
Only code is eligible for AI assistance, subject to the rules above.
Repeated or intentional misuse of AI tools may result in contributor bans.
Maintainers may enforce this policy at their discretion.
Zero-K is maintained by humans.
Every issue, discussion, and pull request is read and reviewed by maintainers who volunteer their time and expertise. Submitting low-effort, unverified, or poorly understood work shifts the burden of validation onto maintainers and is considered disrespectful of that effort.
The primary behaviour this policy seeks to prevent is Vibe Coding, defined as the uncritical generation and submission of AI-produced code without sufficient understanding, verification, or accountability by the contributor. This policy exists to prevent wasted maintainer time and to ensure high standards of code quality and maintainability.
This policy exists to protect maintainers, preserve code quality, and ensure that collaboration remains productive and sustainable.
The restrictions outlined above exist due to repeated misuse of AI by contributors who submit unverified, low-quality, or poorly understood work. The issue is not the tools themselves, but how they are applied.
When used responsibly, transparently, and with proper human oversight, AI can be a valuable productivity aid.
This policy is adapted from the Beyond All Reason adaptation of the AI usage policy published by the Ghostty organisation.
The Zero-K team acknowledges and appreciates the clarity and intent of the original texts.