Git considers voting to determine its LLM policy

Git is reconsidering its policy on contributions written with large language models. Since October 2025, its submission guidelines have warned that contributors may not be able to certify the origin and licensing of substantial AI-generated work, and that generated code and documentation can be misleading or low quality. The guidelines reject work that appears AI-generated or that contributors cannot explain. At the recent Contributors Summit, participants disagreed over whether to relax this stance, citing copyright risks and possible damage to trust. They found little prospect of consensus, prompting a proposal to settle the question by vote.

The draft voting process would let eligible contributors propose policy changes as documentation patches, each backed by another eligible voter. After roughly four weeks of discussion and revision, proposals would go to a two-week ballot. Eligibility might require a merged commit or at least two code reviews, though participants have floated stricter contribution thresholds to prevent people contributing mainly to gain a vote. The proposal does not specify how votes would be counted; secret ballots have been urged to limit pressure from employers, and Debian’s voting system and Condorcet voting are among the options. The process itself remains unsettled, as do voter qualifications and the policy question. Git is weighing a significant governance change that may require broad agreement before it can be used.