rust-lang/rust adopted an LLM policy: "fine to analyze, distill, check, review. Not to create."

Five Rust teams reached consensus on a policy authored by Jynn Nelson, scoped to the monorepo. Allowed: answering questions, analysis, distillation, refinement, checking, suggesting, reviewing, with disclosure required for machine translation and LLM-found bugs. Restricted: creating code changes unless pre-arranged, non-critical, high-quality, well-tested and disclosed; generating public docs; unmarked LLM-written PR descriptions. Soundness-critical changes strongly discouraged even for domain experts. The reasoning is the part to read: polished PRs no longer signal effort or understanding, generation worsens an existing review-bandwidth problem, and LLM-to-LLM copy-paste in review threads wastes maintainer time.