Debian votes on LLM contributions ban

Debian has announced two competing proposals for a general resolution on large language model (LLM) contributions, set for vote after a discussion period ending on 2026-07-24. The proposals represent two sharply different paths forward for the free software project.

Proposal A, authored by Matthias Geiger and seconded by Johannes Schauer Marin Rodrigues, Antoine Le Gonidec, Simon Richter, David Bremner, Pierre-Elliott Bécue, Ian Jackson, and Amin Bandali, seeks to expressly forbid direct contributions to Debian written with LLM or generative AI assistance. The ban would cover Debian source packages, official Debian software (such as lintian), web resources, documentation, and translations. It explicitly would not cover upstream projects already using LLMs, AI-related software itself, or upstream patches and security fixes. The proposal argues that LLM usage is contrary to Debian's Social Contract and reputation for stability. It identifies four core problems: copyright and licensing (LLM output has unclear legal status incompatible with Debian Policy and the DFSG); quality and accuracy (LLMs produce syntactically likely but factually incorrect output, especially problematic in packaging where outputs contain outdated or imaginary copyright and non-functional watch files); community damage (LLM-dependent new contributors do not learn to understand Debian processes and strain volunteer reviewers, risking burnout and loss of expertise); and ethics (LLM companies scrape web content without regard for license or robots.txt conventions, effectively performing a denial-of-service attack that has made parts of Debian's infrastructure unreachable, and consume massive resources while wasting administrator time). The proposal states it was written "organically and without language model assistance."

Proposal B, submitted by Lucas Nussbaum and seconded by Andrey Rahmatullin, Christian Kastner, Anton Gladky, Stefano Zacchiroli, Simon Quigley, Soren Stoutner, and Julian Andres Klode, takes the opposite approach. It proposes to allow AI-assisted contributions under a set of conditions. Contributors must ensure the AI tool's terms do not conflict with distribution and modification under free software licenses; verify license compliance for any pre-existing copyrighted materials in the output; assume full responsibility and accountability for technical merit, security, and licensing; disclose when a significant portion is AI-generated (using methods such as Git trailers like "Generated-By:" or "Assisted-By:"); discuss bulk or autonomously generated changes with the community beforehand (similar to the mass-bug filing process); and avoid tools that transmit sensitive project information to untrusted providers. Proposal B acknowledges the concerns—"about the technical quality and maintainability of such contributions, and their legal status," as well as broader ethical concerns about AI's impact on society, industry, and Free Software, plus environmental and scraping concerns—but argues that with appropriate conditions, contributors can mitigate these risks while benefiting from tools many already find helpful.

The two proposals are equally supported: Proposal A has seven seconders, and Proposal B also has seven, reflecting a genuine division within Debian's developer base on whether AI assistance is fundamentally incompatible with Debian's values or can be governed through disclosure and accountability.

The two proposals represent fundamentally different views on Debian's role and values in the face of generative AI. Proposal A frames LLM usage as incompatible with Debian's core commitment to stability and quality—core tenets of Debian's stated identity and Social Contract. The proposal draws on four concrete concerns: copyright and licensing clarity (essential to Debian Policy and the DFSG), quality and accuracy (particularly acute in packaging, where LLM outputs often contain outdated or imaginary content), community health (arguing that LLM-dependent contributors do not learn the skills needed to sustain the project), and ethics (pointing to AI companies' web scraping practices and resource consumption as harmful to the free software ecosystem). The framing emphasizes that Debian is "a community built on shared interests" where learning and expertise matter.

Proposal B acknowledges the same risks but argues they are manageable through disclosure, accountability, and careful process. It recognizes that "many Debian contributors find AI tools helpful" and that benefits and risks both exist. By requiring contributors to understand their changes and remain accountable, to verify license compliance, and to disclose AI use, Proposal B aims to preserve Debian's quality standards while permitting a tool that has become common in development. The approach leaves room for the project's position to "evolve as time passes" without further general resolutions, framing the decision as clarification rather than a final boundary.

Both proposals have equal support within Debian's leadership—each is seconded by exactly seven developers—indicating genuine disagreement on whether AI assistance is fundamentally at odds with Debian's values or merely requires better governance. The vote will signal whether Debian sees AI as a tool that can be integrated with appropriate guardrails, or as a practice incompatible with the stability and community learning that define the distribution.

No comments yet. Be the first to share your thoughts!