Debian’s General Resolution to ban AI-assisted contributions opened its official discussion period on July 24, 2026. This is not a mailing list argument — it is a formal vote. If it passes, using Copilot, Claude Code, or any generative AI tool to write Debian packaging, documentation, or native project code becomes a policy violation. The timing is hard to miss: Linus Torvalds told AI critics to fork it ten days earlier.
What the GR Actually Proposes
The proposal, filed by Matthias Geiger and Jesse Rhodes, targets direct Debian contributions: source packages, native Debian software like lintian and apt, packaging work, documentation, translations, and official communications.
The headlines are oversimplifying it. The ban explicitly excludes upstream projects that happen to use AI during their own development, AI-related packages, and security patches from upstream. You can package an application written by an AI in Debian. You cannot use an AI to write the Debian-specific packaging around it.
There is also a competing resolution on the table — one that takes a different approach entirely. Instead of prohibition, it requires disclosure: contributors can use AI assistance but must take full responsibility for quality, security, and licensing compliance, must be able to understand and explain everything they submit, and must flag AI assistance in commit metadata.
The Four Concerns Behind the Ban
The proposal cites four rationales: copyright uncertainty over LLM-generated code, technical quality risks from hallucinated APIs and subtle bugs, impact on Debian’s volunteer community dynamic, and broader ethics around generative AI training data and energy use.
The copyright argument is the weakest. Debian contributors using Copilot to write a packaging script are not liable for what OpenAI or Anthropic trained on — that legal question sits with the vendors, not the developers. The EU AI Act’s August 2 disclosure requirements actually align more closely with the disclosure option than with the ban.
The quality concern is real but already handled. Debian’s review process exists precisely to catch bad code, regardless of origin. The disclosure option gives reviewers the additional context they need to weigh contributions appropriately.
Enforcement Is the Real Problem
No tool reliably detects LLM-generated code at scale. The ban depends entirely on contributor self-reporting. AI-assisted covers a wide range — a single autocomplete suggestion, a full function generated by Claude, a complete file rewrite via Cursor. Where does assistance end?
A policy that cannot be enforced is not a policy — it is a statement of values. That may be what Geiger is after, and values statements are legitimate. But the disclosure option achieves the same thing more honestly: it does not pretend the behavior can be policed, it asks for transparency instead.
The Open-Source AI Governance Reckoning
Debian is not deciding in isolation. In the last four weeks, the open-source community has been answering the same question through very different lenses:
- The Godot Foundation banned AI-generated contributions (June 30)
- Linus Torvalds told AI critics to fork Linux (July 15), cementing a policy of human accountability rather than tool restrictions
- Codeberg voted 358 to 144 to ban repositories that mostly consist of AI-generated code (July 22) — driven primarily by storage and CI/CD cost sustainability
- Debian filed this GR (July 22), discussion opened July 24
Each project is answering the same question differently based on its own structure and constraints. Codeberg’s ban makes sense for a donation-funded hosting platform drowning in vibe-coded repos. Linux’s accountability model makes sense for a project with rigorous patch review. Debian’s situation is more like Linux than Codeberg — the quality gate is the review process, not the tool used to write the first draft.
The Case Against the Ban
A flat ban on AI-assisted contributions would make Debian the most restrictive major Linux distribution by policy. Ubuntu, which derives from Debian, has not signaled any similar restriction. Fedora, Arch, and NixOS have not moved in this direction. Debian already struggles with package maintenance backlogs — a policy that further deters contributors risks accelerating that problem rather than solving it.
The disclosure option is the better answer. It preserves human accountability, gives reviewers information they can act on, aligns with the EU AI Act’s transparency direction, and does not require an unenforceable detection mechanism. It treats AI as what it is: a tool, not an author.
The GR discussion period is open now. If you are a Debian developer, this is the moment to contribute your view to the debian-project mailing list. If you use Debian packages or contribute documentation, the outcome affects your workflow directly.
