Debian 投票通过“负责任使用生成式 AI”决议

The results of the Debian general-resolution vote on the use of large language models have been posted; the winner is choice 5: Responsible Use of Generative AI.

Debian neither endorses nor prohibits the use of generative AI tools in the development, maintenance, or documentation of software, packaging, documentation, and other media published within the Debian Project. We recognize that such tools can substantially improve the productivity of contributors when used responsibly, allowing volunteers to spend more of their limited time on work that requires technical expertise, judgment, review, and collaboration.

The Debian Project nevertheless expects that all contributions submitted to Debian, regardless of how and with which tools they were produced, satisfy the same standards of quality, correctness, maintainability, and legal compliance. The use of a generative AI tool does not diminish the contributor's responsibility for the work they submit. Contributors are expected to understand, review, test, and, where appropriate, modify AI-assisted output before incorporating it into Debian.

to post comments

Inevitable

Posted Aug 29, 2026 14:03 UTC (Sat) by dskoll (subscriber, #1630) [Link] (1 responses)

I'm a bit sad it went that way, but I guess it was inevitable. Debian is too large a project to be able to meaningfully enforce a stronger stance against LLMs. So I think they realized it's pointless to make rules that you can't enforce.

Inevitable

Posted Aug 29, 2026 15:06 UTC (Sat) by MarcB (subscriber, #101804) [Link]

I, for one, am very glad. In particular the fact that the two full-Anti-AI proposals (1&3) were even below "None of the above" is very good sign. Had one of those won, the attempts to enforce them would have ruined Debian.

Extremist positions soundly defeated

Posted Aug 29, 2026 14:20 UTC (Sat) by bluca (subscriber, #118303) [Link] (3 responses)

It might not be obvious due to the complexity of the voting system, but it is definitely worth noting that not only the sensible and rational position won, but the options that were set up with the implicit (or by reading some emails from some proponents, quite explicit actually) purpose of setting up witch hunts to actively punish anyone who dared not toe the extremely vague anti-AI line (option 1 to change social contract, option 3 to change code of conduct, both of which imply *expelling* any dissenter) were soundly defeated and rejected: they (and only they) both failed to beat the "None of the Above" option, which is this voting system's way to say "no, absolutely not, go away".

The rest of the options cleared that line and then it was a (complicated) matter of which one was the "least unfavored" among all voters, so to speak.

A handy graph that someone else prepared makes the result a lot more digestible: https://people.debian.org/~lucas/gr-2026-002/results.png

Extremist positions soundly defeated

Posted Aug 29, 2026 14:41 UTC (Sat) by mhvk (subscriber, #86585) [Link] (2 responses)

It depends what you mean by "soundly" - if I read that graph correctly, about 30% of the voters preferred the no-AI options over all the others. To me, it looks like "status quo" won, which is perhaps appropriate in times of rapid change.

Personally, I hope that truly open locally run models, perhaps specialized to coding, become sufficiently good that one can do without the large, closed-source ones (which to me are very problematic for lots of reasons; it doesn't help to come from a country that will be under water -- the Greenland icesheet melting completely is nearly unavoidable by now, and 6 meters of sea level rise is more than one can build dykes for; why these data centers are not forced to rely on renewable energy is truly beyond me).

Extremist positions soundly defeated

Posted Aug 29, 2026 14:45 UTC (Sat) by bluca (subscriber, #118303) [Link]

> It depends what you mean by "soundly"

With this voting system, failing to beat "NOTA", as mentioned, means the option is soundly rejected even before considering relative preferences

Extremist positions soundly defeated

Posted Aug 29, 2026 15:20 UTC (Sat) by MarcB (subscriber, #101804) [Link]

> Personally, I hope that truly open locally run models, perhaps specialized to coding, become sufficiently good that one can do without the large, closed-source ones...

In theory, they are. The problem is that the absurd investments into AI hardware have made it unaffordable. 64-128 GiB+ GPUs at home would be absolutely realistic without this cost explosion.

> ...the Greenland icesheet melting completely is nearly unavoidable by now...

Technically not, but I also believe we will not avoid it. But the reason for that is not AI. Its share of human energy consumption is somewhere around 0.2%, likely lower. And AI would be very amenable to using renewal energies - if any relevant government cared enough to enforce its usage.

A growing disconnect

Posted Aug 29, 2026 14:44 UTC (Sat) by aimannajjar (subscriber, #184277) [Link]

Ed Zitron's interviews gaining 5million views overnight on YouTube, while the people of free software embrace AI unconditionally is the wildest ironies in history.

It will be interesting to watch what happens next in terms of quality and review processes as well as young engineers skills. I genuinely think it's good to have diversity in AI adoption so we can compare and contrast the net results across projects in the long term.

Maintainer opinions?

Posted Aug 29, 2026 15:13 UTC (Sat) by jpeisach (subscriber, #181966) [Link] (2 responses)

What if a package maintainer feels that they wish to reject a PR because the description is written by a LLM? Are they allowed to do that?

Maintainer opinions?

Posted Aug 29, 2026 15:28 UTC (Sat) by hlieberman (subscriber, #123867) [Link]

Speaking only as a DD, but not officially on behalf of the project:

As long as you are being respectful and following the other requirements of the Code of Conduct, the only group with the ability to force a maintainer to accept a patch would be the Technical Committee. That's a high bar, and generally the only things that go in front of them are issues around major packages or mass-filings, or when two maintainers heatedly disagree.

As a maintainer of a package, I am responsible for what goes into that package. Another maintainer can disagree with my decisions, but absent a ruling of the TC, they can't override my decision*.

*: Technically, a developer _could_ just ship an update with their change, but doing so over the objections of the maintainer would almost certainly end up getting you in a world of trouble.

Maintainer opinions?

Posted Aug 29, 2026 15:30 UTC (Sat) by cen (subscriber, #170575) [Link]

AFAIK a Debian maintainer generally has broad discretion over which patches to accept for any reason, including AI. But they cannot use that discretion to refuse to implement a binding Debian project/Technical Committee decision or an applicable Debian Policy requirement.

Maintainability

Posted Aug 29, 2026 15:40 UTC (Sat) by akselmo (subscriber, #174307) [Link] (4 responses)

I have a feeling this will cause maintenance and stability issues further down the line, as people rarely are "responsible" with

these tools. I and my servers hope I am wrong.

Maintainability

Posted Aug 29, 2026 16:41 UTC (Sat) by dilinger (subscriber, #2867) [Link] (3 responses)

It really was the worst possible option. As Bunk said (https://lists.debian.org/debian-vote/2026/08/msg00367.html), lack of disclosure is by no means responsible. The result leaves it at the status quo, eg the current wild west. It feels to me like the project as a whole voted to kick the can down the road a few years (and I've seen a few people who voted for the option say as much).

Maintainability

Posted Aug 29, 2026 16:54 UTC (Sat) by MarcB (subscriber, #101804) [Link] (2 responses)