The Rust Project is evaluating a formal policy to strictly limit the use of large language models in contributions to the main rust-lang/rust repository.
The proposal, now an open PR in the Rust Forge repository, includes an “LLM Usage Policy” for rust-lang/rust. It does not ban AI tools entirely, but distinguishes between private use of LLMs and their use in creating public project content.
The draft policy permits LLMs for answering questions, analyzing code, summarizing comments for private use, private code or prose review, and suggesting solutions. However, it stresses that LLMs should be used to “write better, not faster,” and must not replace human understanding, authorship, or review.
The proposed rules would prohibit comments, issue bodies, and pull request descriptions from personal GitHub accounts if originally created by an LLM. The policy would also ban documentation generated by LLMs, including non-trivial source comments, doc comments, safety comments, multi-paragraph source comments, and compiler diagnostic messages.
The draft also addresses the review process. Under the proposed policy, LLM reviews would be advisory only and could not determine whether to merge or reject changes. Human review would remain required where project rules specify, and LLM reviews would not replace author self-review.
Interestingly, the draft still allows limited LLM use with disclosure, such as machine translation, trivial code or prose changes, and bug discovery verified by the contributor. LLM review bots may be used if they follow maintainer-approved rules, operate from a clearly marked separate GitHub account, and their comments do not block pull requests unless endorsed by a human reviewer.
Additionally, a separate experimental path would permit some LLM-generated code changes under strict conditions. These changes must be pre-arranged with a reviewer, non-critical, high-quality, well-tested, and thoroughly reviewed. The policy explicitly states the project does not accept “vibe-coded” pull requests that lower codebase quality.
The proposal defines a narrow scope, applying only to rust-lang/rust and teams that ratify it, such as compiler, libs, types, rustdoc, bootstrap, and their subteams. Other Rust organization repositories, submodules, subtrees, crates.io dependencies, and teams like lang and edition may set their own policies.
If adopted, the policy would provide Rust with one of the clearest public rule sets for AI-assisted work in a major open-source language project.
