Rust bans autonomous AI contributions, lets reviewers close PRs

Rust banned autonomous AI agents from submitting code on August 5, per a policy posted to the project's blog. LLMs can review, analyze, refine, and suggest. They can't create.

The rule targets a specific cost transfer. A pull request that takes an author 90 seconds to generate can cost a reviewer 40 minutes, against a backlog of 1,281 open PRs. Reviewers can now close a noncompliant PR without giving a reason.

The specifics go further than the headline ban. LLM-generated code is barred from public documentation. Mechanically generated replies to review comments are barred too. Soundness-critical changes require a human domain expert, full stop. Disclosure is mandatory for machine translation, trivial changes, bug discovery, and any LLM-generated public text. A translated comment or a "small" AI-assisted fix now needs a label too.

The same instinct showed up somewhere smaller. Sebastian Pipping, who maintains libexpat, said the City of Munich's Open Source Sabbatical program is funding him for up to six months starting August 1. The work prioritizes five open CVEs and XML 1.0r5 support. Buried in his announcement: unvalidated AI slop submissions still won't be appreciated. A municipal government is now paying to keep a library alive that ships inside nearly everything. Its maintainer opened with the same objection Rust just wrote into policy.

I generate code with agents most days on my own projects. Rust's line matches the one I already hold myself to: understand every line well enough to defend it in review, or don't ship it. Pointing an agent at an issue and forwarding the diff is a different activity than that, and Rust just made the difference official.

The ban on autonomous agents will get cited the most. The clause that actually changes daily behavior is the silent close, because it means reviewer time now outranks a contributor's need for an explanation.