Debian Votes to Allow AI Code. The Submitter Is Still Responsible for All of It.
Debian has settled its AI debate, and the answer is neither a ban nor a green light.
Developers voted on eight competing general-resolution proposals over two weeks. The winner, announced this week, is Choice 5: “Responsible Use of Generative AI.” It passed via Schulze voting — a preferential method used to rank outcomes when no single option commands a majority.
The outcome mirrors the Linux kernel’s position: tools are tools, submitters own their output.
What the Resolution Actually Says
The full text is precise about what Debian is and is not doing. The project “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.”
The quality standards — correctness, maintainability, legal compliance — remain identical regardless of how a contribution was produced. The resolution is explicit: “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.” There is no mandatory disclosure requirement. But if the code breaks, the person who submitted it owns the breakage.
Why This Matters Beyond Debian
Debian is not a minor distribution. It underlies Ubuntu, and through Ubuntu, a significant portion of the server infrastructure running the internet. Its governance decisions propagate downstream, and its maintainer norms influence open-source culture broadly.
The eight-option ballot included positions ranging from an outright prohibition on AI-generated contributions to permissive frameworks with lighter obligations. The vote landed in the middle: permissive on tool use, strict on accountability.
That framing cuts against two common positions in the AI code debate. It rejects the view that AI-assisted code is inherently problematic — Debian explicitly says AI 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.” But it also rejects any implication that submitting AI-generated code without review is acceptable.
The position is structurally identical to the Linux kernel’s existing stance. Linus Torvalds has said he does not particularly care how code is written, only whether it works and passes review.
The Open-Source AI Policy Landscape
Several high-profile open-source projects have moved in the opposite direction over the past year, adding explicit prohibitions or heavy disclosure requirements for AI-generated contributions. Debian’s vote provides a counterpoint: a project of significant institutional weight that chose not to treat AI as categorically different from any other tool.
The vote does not resolve ongoing questions about license compliance when AI training data includes copyrighted code, or about how maintainers should audit AI-assisted patches at scale. Those are left to individual contributor judgment — which is, in practice, how most of Debian’s quality standards have always been enforced.
The resolution is effective immediately. Eight proposals went in. One came out. The Debian project’s position on AI is now on record.