Five teams within the Rust project have adopted a new LLM usage policy for the rust-lang/rust monorepo. The policy isn't an official project-wide stance on AI — it's a targeted response to problems moderators were already handling informally.
Why it was created:
Polished, well-tested PRs used to signal genuine effort and understanding. LLMs broke that signal — a slick PR no longer means the author understands the code, or that they'll stick around as a contributor. Easier code generation is worsening an already severe reviewer-bandwidth problem (1,281 open PRs at time of writing). Contributors copy-pasting reviewer comments into an LLM and pasting the reply back is described as a waste of time and a breach of trust.
Core rule: "It's fine to use LLMs to answer questions, analyze, distill, refine, check, suggest, review. But not to create."
Key provisions:
LLM-generated content must always be disclosed if posted publicly (PRs, issues, comments); undisclosed private use is fine. LLM-generated code is allowed only if pre-arranged, non-critical, well-tested, and disclosed — held to a higher bar than human code, not a lower one. LLMs cannot touch soundness-critical code unless the author is already a domain expert. Reviewers may close policy-violating PRs without justification and direct authors to a dedicated mentoring channel. Harassing someone for using an LLM is explicitly against the rules. The policy is designed to be easy to revise, since Rust governance works by consensus, not top-down bans.
The policy currently applies only to the rust-lang/rust repo, though the author hopes for a broader, project-wide policy eventually.
