Rust project publishes LLM policy: permits assistance, bans authorship; enforces disclosure for PRs

The Rust project published an official LLM policy on August 5, 2026, addressing months of heated internal debate triggered by low-effort 'slop' PRs and maintainer burden. The policy, adopted by five core teams, distinguishes between permissible and prohibited LLM use. Allowed uses include asking LLMs to analyze or summarize existing code, using them privately for self-review before posting, and querying them for education—only authors need see the output. Prohibited uses include LLM-generated code in PRs, LLM output in public comments or issue descriptions, and LLM reviews substituting for human review or self-review. Policies must be written for humans first, then optionally summarized for machines.

Disclosure is mandatory for non-trivial LLM use in public contributions. Any LLM-assisted PR must be clearly marked, and reviewers are not required to review them if they choose not to. Contributors who lie about LLM use face Code of Conduct violations, not merely policy violations. An experimental exception allows LLM-authored code under strict conditions: prior reviewer solicitation, non-safety-critical changes, well-tested and well-reviewed contributions, and both author and reviewer must fully understand the code. Such PRs are tagged 'ai-assisted' and posted to a private Zulip channel to track outcomes.

For maintainers and open-source projects: Rust's policy reflects a structural shift in how projects govern automation. The distinction—assist but don't author—is pragmatic: LLMs excel at code review aids and explanation, but lower effort signals on polished PRs harm community trust and review bandwidth. Rust has 1,281 open PRs; low-effort contributions exacerbate existing review backlogs. The COC-level enforcement of disclosure (vs. mere policy violation) signals seriousness about transparency. Expect other large projects to adopt similar governance frameworks as LLM tooling becomes standard in developer workflows. The unresolved tension: whether restricting LLM authorship now forecloses future contributors for whom LLMs are their primary coding tools.