skip to content

AI-assisted contribution — where we stand

Some open-source projects ban AI-assisted contributions outright. Gentoo, NetBSD and Servo have adopted full, written bans, and QEMU is discussing walking one back. Others — including some of the largest projects in existence: the Linux kernel, Kubernetes, LLVM, CPython, Fedora and Django — have adopted policies that permit AI-assisted contribution, provided the human stays accountable for the result.

What those projects ask about disclosure varies, and it is worth being precise. Kubernetes and Django require it. The kernel asks for an Assisted-by tag. Fedora wants it when a significant part of the work comes from a tool unchanged. CPython appreciates it without requiring it, and LLVM's policy turns on human review, and on limits to where agents may work, rather than on disclosure. Kubernetes, and a separate unmerged Rust proposal, go further and forbid using an agent to answer reviewers on your behalf.

Read across the adopted policies and one question recurs, in different words: will you, the engineer, stand behind what you submitted? Not the tool — the outcome, and your ability to defend it under review.

The data supports a measured version of this. Agent-authored pull requests merge at high rates, between 43% and 87% depending on the agent and the task, and the same factors decide agent and human pull requests alike: tests passing, reviewer engagement, no duplicated work.

What we do about it

Before you can claim an issue, the CLI reads that repository's own CONTRIBUTING, pull-request template and AGENTS files, and classifies what they say.

  • Prohibited — we refuse to record the claim. You can continue only by stating outright that you will hand-write the work yourself.
  • Disclosure required — you are told before you start, along with the wording the project asks for.
  • Unreadable — if the docs are rate-limited or missing, we say so and tell you to read them yourself. We never treat "couldn't check" as "no policy."

The check runs before any code is written, not after a pull request exists. We treat it as a precondition.

Related: Contribute · Claims · Proof of Work