QEMU eyes end to blanket ban on AI-generated contributions

QEMU is considering a full reversal of its ban on AI-generated contributions, replacing an absolute prohibition with disclosure rules and clear limits on what machines may write.

Maintainer Paolo Bonzini posted a request-for-comments series on the QEMU development list that would allow large language models for code, tests, documentation, and research. Contributors would mark such work with an AI-used-for trailer ahead of their Signed-off-by line and would still certify origin under the usual DCO. The prior policy rejected anything believed to include or derive from AI output. Bonzini argues that usable LLM output has made a hard ban difficult to justify or enforce, and that high-quality AI-assisted patches have already landed despite it.

The draft is not open season. Cover letters, commit messages, replies to reviewers, and issue text must remain human-written, with only light copy editing allowed. Heavily assisted or large submissions, such as new device models, would need a maintainer’s prior agreement to review them. Maintainers keep their own judgment on what to accept, with the stated goal of avoiding burnout while heeding Software Freedom Conservancy advice not to shut out either camp.

Bonzini also proposes project instructions for coding agents so they flag requirements early, refuse violating requests, keep humans in design choices, and hand over drafts rather than finished submission text. The approach draws on Conservancy guidance and policies from projects including Rust and GStreamer. Early list replies have been constructive, including review tags on parts of the series and clarification that copy editing may help non-native English writers without letting models author substantive community text.