Rust Adopts Formal LLM Policy for Its Main Repository
The Rust project announces new rules governing the use of AI tools in contributions.
1Rust's LLM Policy Announced
Five core teams in the Rust project have adopted a formal policy governing how large language models (LLMs) can be used when contributing to the main rust-lang/rust repository. The policy was announced on the Inside Rust blog on August 5, 2026. Authored by Jynn Nelson, this step formalizes what the project previously described as an unwritten 'wild west' approach to moderation. The policy applies only to the rust-lang/rust repo and the teams that ratified it.
2LLMs Accepted as Tools, Not Substitutes
The policy encourages using LLMs as thinking tools while strictly rejecting their use as replacements for thought. The document summarizes itself in one sentence: 'It’s fine to use LLMs to answer questions, analyze, distill, refine, check, suggest, review. But not to create.' This approach draws clear boundaries on how and when LLMs can be used in the contribution process.
3Three-Tier Usage Policy
The policy divides LLM usage into three tiers. Fully permitted: private uses where only the contributor sees the output (asking questions about the codebase, summarizing a thread, reviewing one's own code). Permitted with mandatory disclosure: machine translation, trivial changes like typo fixes, LLM-assisted bug discovery, and clearly labeled LLM review bots running from separate accounts. Banned: LLM-created comments, documentation, and compiler diagnostics; any process requiring an LLM to execute; and treating an LLM review as sufficient to merge or reject a change.
4Strict Enforcement Mechanisms
The policy's toughest aspect lies in its enforcement design. Deliberately misrepresenting LLM use is treated as a violation of the project's Code of Conduct—equivalent to harassment—resulting in warnings and, for repeat offenses, bans. The document explicitly states many clauses are unenforceable in practice, calling this intentional: 'Our goal isn't to catch every violation... Instead, it's to remove plausible deniability: forcing a choice between following the policy and intentionally violating it.'
5Experiment for LLM-Written Code
Code written by LLMs isn't banned outright but confined to an experiment with strict conditions. Changes must be pre-arranged with a named reviewer, non-critical to compiler soundness, well-tested, well-reviewed, and always disclosed. New contributors can't open an LLM-created pull request without first securing a reviewer. If no test suite exists for modified code, the author must write one or close the PR. No exceptions.
6Circuit Breaker for the Experiment
The experiment carries its own circuit breaker. If over half of PRs merged in a six-week window are LLM-created, merges stop until the share drops below 50%, with a minimum 10-day cooldown. The window aligns with Rust's six-week release cycle. All such PRs carry an 'ai-assisted' label and are posted to a private Zulip channel for data collection—tracking whether LLM-assisted contributors are learning, returning, and producing useful work—not for gatekeeping.
7Three Pressures Behind the Policy
Nelson's announcement describes three pressures pushing teams from informal moderation to written rules. Polished PRs no longer signal effort or understanding, eroding trust signals the review culture relies on. Cheaper code generation worsens existing review bandwidth shortages: the repo currently has 1,281 open PRs, and the scarce resource was always reviewer judgment, not code. Contributors pasting review comments into LLMs and back are, in Nelson's words, 'wasting everyone's time and breaking the assumption that a reviewer is talking to a person.'
8A Genuine Split Within the Project
The backdrop is a real division within the project. The policy's motivation section states there's no consensus in Rust—'and likely never will be'—about when AI tools are acceptable, with members ranging from daily users to those rejecting any use. Thus the document is designed to change: major revisions require each ratifying team's sign-off, and the policy can be dissolved by those teams or overridden by a project-wide LLM committee the leadership council is considering.
Original article: Unite.AI — Rust, Ana Deposunda LLM Kullanımı İçin Resmi Politika Belirledi