GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

Debian Puts AI Policy to an 8-Way Vote — From Social Contract Ban to Climate Refusal

Debian opened its General Resolution vote on LLM usage today. Voting runs through August 28. Only Debian Developers can participate, ranked via Condorcet ballot. The result will be binding project policy.

This is the third attempt at a formal resolution. An initial GR in 2025 was withdrawn before it reached a vote. A second attempt earlier in 2026 dissolved without agreement. The current ballot — with a discussion period from July 23 to August 13, 8 competing proposals, and a fixed voting window — is the first to actually start voting.

The 8 Choices

Choice 1: Ban LLM contributions from Debian via Social Contract Proposed by Matthias Geiger. Amends the Social Contract to explicitly prohibit contributions written with LLM assistance. Scope includes source packages, Lintian, web resources, documentation, translations, and official communications. Upstream packages that happen to use LLMs are exempt. Rationale: copyright ambiguity, quality concerns, community erosion.

Choice 2: Allow AI-Assisted Contributions with conditions Proposed by former Debian Project Leader Lucas Nussbaum. Six conditions: legal compatibility with Debian’s licensing, attribution for third-party code in AI output, full contributor responsibility for quality, disclosure when AI assistance is significant (via Git trailers like Generated-By: or Assisted-By:), no uploading of confidential material to AI services. This is a policy document under Constitution 4.1(5), not a Social Contract change — meaning it can be updated without a future GR.

Choice 3: Reject LLMs as far as practical, update Code of Conduct Authored by Ian Jackson. A strong rejection stance that stops short of a technical ban. Requires disclosure of all LLM use in Debian work. Individual maintainers and teams may impose complete bans on their own packages. Violations treated as Code of Conduct violations. Notably: direct communications to people (bug reports, mailing lists, Salsa discussions, Planet Debian posts) must be written without LLM assistance.

Choice 4: Accept AI contributions for Debian-specific work Covers Debian websites, packaging, applications, and resources. Upstream excluded. Requires DFSG compliance, contributor understanding and ability to defend the output, Debian-infrastructure-hosted uploads only, no cloud AI for confidential project information.

Choice 5: Responsible Use of Generative AI Neither endorses nor prohibits. Requires contributors to understand and review AI output before submitting. Encourages but does not mandate disclosure. Explicitly declines to resolve unsettled legal questions about AI copyright in the resolution text.

Choice 6: A cautious approach to generative AI A middle path between conditional acceptance and active discouragement. Specific text not fully reproduced in public coverage; positioned between Choices 4 and 5 in liberality.

Choice 7: Debian is created by humans Disallows the output of generative AI as a direct contribution to Debian, but does not restrict Debian contributors from using AI tools to inform their own work. The distinction is between AI-generated output entering the project versus humans reading AI output and writing their own code.

Choice 8: Avoid the use of LLM: climate destruction is a deal breaker Encourages contributors to avoid LLM use where practical, citing the energy cost of large-scale AI training and inference. Does not mandate; frames the ask around environmental responsibility rather than technical or legal grounds.

Choice 9: None of the above

Why This Is Harder Than It Sounds

Debian’s social structure makes this genuinely contested. The project uses a Condorcet system with ranked preferences, meaning voters can express nuanced positions. Nussbaum’s Choice 2 draws backing from contributors who use AI tools daily. Jackson’s Choice 3 draws backing from those who argue that allowing LLMs gradually destroys the mentorship pipeline that brings new contributors into the project. Geiger’s outright ban (Choice 1) has eight confirmed seconds — enough to put it on the ballot — and draws on DFSG copyright concerns that have no clean resolution under current law.

The climate argument (Choice 8) is unusual but not marginal. Debian has historically taken principled positions on free software that precede mainstream adoption. Several developers have argued that AI’s energy footprint places it in the same category as proprietary software — a fundamental values incompatibility, not just a policy inconvenience.

What actually passes depends on the Condorcet count. In ranked voting, the winner is the option that beats all others head-to-head. A strongly disliked Choice 1 could lose even if it gets the most first-preference votes. A broadly acceptable Choice 5 could win without many top-ranked placements.

The result will be published after August 28. Whichever option clears Condorcet and the required supermajority threshold becomes binding policy for all Debian contributions.