Mesa will allow AI-generated code if author understands it (www.phoronix.com)

Mesa, the open-source graphics driver project, has published contributor guidance that permits using AI tools to generate or audit code — but only if the human author truly understands and takes responsibility for the changes. The policy followed a recent incident where an independent contributor used a “GPT5‑HIGH” AI code audit to suggest optimizations to RADV’s command buffer code; after implementing and reviewing the suggestions, the author reported roughly a 1% performance uplift in Cyberpunk 2077 and Total War: Troy. The Mesa community pushed back when contributors posted raw AI output, since maintainers would otherwise have to decipher, validate and patch uncertain changes themselves. The new guidance emphasizes reviewability and accountability: contributors must know Git, explain why their changes benefit the project, and produce patch-level edits that can be properly reviewed — tooling is up to the author, but tools don’t replace comprehension. Technically this means AI can be used as a productivity or audit aid, but every change still needs human validation for correctness, performance, and safety; crude paste-ins of AI suggestions are discouraged. Mesa maintainers also note the policy is still evolving and discussion will continue, reflecting broader open-source concerns about correctness, licensing, and maintainability when AI-assisted code becomes common.