Summary: Debian opened a General Resolution titled “LLM usage in Debian” for discussion on July 23, with voting scheduled from August 15 through August 28. Eight ranked proposals range from banning LLM-assisted contributions to requiring disclosure or welcoming them. The project-wide vote matters because its charter-level result can set a formal policy for contributions and reviewer workload.
Debian has spent more than a year arguing over a single question, and it is now reaching for its heaviest formal instrument to settle it: may content produced with LLM assistance enter the distribution? On July 23, a General Resolution titled “LLM usage in Debian” entered its discussion period, with the vote scheduled to run from August 15 through August 28. A General Resolution sits at the charter tier of the project’s governance and is invoked only rarely — the best-known case was the fight over the init system — and this time the subject in dispute is generative AI.
The Ballot: Eight Proposals and a “None of the Above”
Voters are not being asked to pick between two opposing poles. Eight proposals share the ballot with a “None of the above” option, and the positions cover the whole range: outright prohibition at one end, mandatory disclosure in the middle, an explicit welcome at the other. Counting follows the Condorcet (Schwartz) method, which has voters rank their preferences instead of forcing them onto a single side. The proposals were collected on the debian-vote mailing list, and among the drafters are some of the project’s most senior developers.
Three Proposals in Detail
The first proposal, drafted by Matthias Geiger, takes the hardest line on the ballot: it amends the Social Contract so that any contribution written with LLM assistance is banned outright. Package source code, Debian’s own software, web resources, documentation and translations, and official communication all fall inside that scope. Touching the Social Contract is what raises the bar to a 3:1 supermajority. Four concerns are set out in its rationale: a copyright status that cannot be established, quality and correctness that cannot be vouched for, reviewers pushed into burnout, and the DoS-like strain LLM crawlers place on Debian’s infrastructure.
Drafted by Lucas Nussbaum, the second proposal would let AI-assisted contributions through, but only with conditions: the terms attached to the tool in use must not conflict with Debian’s purpose; the contributor bears full responsibility; disclosure becomes mandatory once AI-generated material makes up a significant share, for instance through a Generated-By: trailer recorded in the git history; bulk or automated changes have to be discussed on the lists first; and confidential data may not be handed to untrusted AI providers.
The third proposal, from Ian Jackson, asks contributors to shun LLMs so far as is practical and attaches a supplement to the Code of Conduct: anything addressed to a human must be drafted by a human alone; any use of an LLM must be disclosed; bans adopted by individual projects must be respected; and a breach of any of these is handled as a violation of the Code of Conduct.
Five further options occupy the ground between proposal three and the permissive end of the ballot, among them a middle-ground choice titled “Responsible Use of Generative AI.” The spread is deliberate. Debian’s process forces every position onto the record one by one rather than flattening the argument into a binary, and the Condorcet count then identifies whichever option defeats every other in head-to-head comparison. On a question where almost everyone has a “but” attached — yes, but ban it; no, but disclose it — a ranked ballot serves better than a yes/no referendum.
Why This Matters So Much to Debian
On its face the dispute looks philosophical. Reviewer bandwidth is what actually drives it. Volunteers maintain Debian’s packages, and low-quality AI-generated bug reports and patches land directly on their desks — proposal one goes so far as to cite cases in other projects where AI contributions pushed maintainers out. Those on the open side counter that AI assistance is already ordinary practice, that pretending otherwise helps nobody, and that the better move is to state the responsibility and disclosure rules plainly. The proposal texts lean on plenty of precedent: the GNOME Loupe contribution dispute, Gentoo’s AI policy, Codeberg’s related changes, and the AI usage policies GCC and Rust have each published.
What sets Debian’s answer apart is the instrument. Elsewhere the matter gets settled in a maintainer meeting or a README; Debian settles it in its constitution, where the outcome binds the entire project and can be undone only by another vote. Slower, noisier, and far harder to quietly reverse — which is precisely why the wider open-source world is following the discussion thread.
Timetable and Passing Thresholds
Discussion stays open through August 13; the vote begins August 15 and ends August 28. Since it amends the Social Contract, proposal one requires a 3:1 supermajority, while the remaining options pass on a simple majority; every option also has to clear Debian’s quorum requirements against “None of the Above.” New amendments can still arrive while the discussion window is open — part of the reason the ballot reached eight choices in the first place.
However the vote lands, Debian will be the first major distribution to set a formal, project-wide position on LLM contributions through its own governance process. Open-source communities still stuck at the “each maintainer decides” stage can take the ballot design and the debate record as a reference already assembled for them.
Sources
- Debian — the General Resolution page for LLM usage in Debian
- Proposal-one — barring LLM contributions by amending the Social Contract
- Proposal-two — AI-assisted contributions, allowed under conditions
- Proposal-three — reject LLMs, revise the Code of Conduct