Debian Launches Community Poll to Determine AI Tool Guidelines for Project Contributions

Debian is asking its developer community to weigh in on the use of AI tools, ranging from unrestricted adoption to outright prohibition.

The second Debian vote of the year, titled General Resolution: LLM usage in Debian, is currently open. Debian Project Lead Sruthi Chandran, who was elected in March in the project’s first 2026 poll, has extended the voting period by an additional week.

Given Debian’s massive scale—version 13 ships with 69,830 packages occupying roughly 403 GB and comprising more than 1.46 billion lines of code—the poll is necessarily intricate.

The poll spans over 5,000 words and includes eight distinct proposals, each backed by between six and 17 developers. They are as follows:

1. Ban LLM-generated contributions to Debian under the Social Contract.

2. Permit AI-assisted contributions subject to specified conditions.

3. Reject LLMs in practice and revise the Code of Conduct accordingly.

4. Accept AI-generated work for Debian-specific tasks.

5. Advocate responsible use of generative AI.

6. Adopt a cautious, measured approach to generative AI.

7. Maintain that Debian is authored by human contributors.

8. Avoid LLM usage due to its environmental impact.

The poll notes:

“Proposal A needs a 3:1 majority, the other proposals need a simple majority.”

Each proposal varies in structure; some include explanations, bullet points, and summaries. Only recognized Debian developers may cast a vote, but it is hoped that the detailed text will not discourage participation and that voters will carefully identify the option that best aligns with their stance.

Proposal A is among the longest. Its “preamble” states:

“This proposal aims to expressly forbid any contributions to Debian written with the use or assistance of large language models (LLMs) or other generative AI tools.”

Proposal B would permit LLM-based contributions, provided they satisfy six criteria: legal compliance, licensing and attribution, accountability, disclosure, prior discussion of bulk or automated changes, and confidentiality/privacy.

Proposal C calls for a blanket ban, summarizing its stance as: “Reject LLMs (generative AI) as far as practical.” It is supported by Debian veteran Ian Jackson, the author of dpkg and operator of Chiark.

Proposal D seeks a pragmatic middle ground, acknowledging that such tools are already in use and unlikely to disappear. Rather than imposing a ban, which could be counterproductive and difficult to enforce, it places responsibility on contributors and outlines the following guidelines.

Proposal E similarly emphasizes that each contributor remains accountable for the technical quality, legal compliance, and appropriateness of their submissions.

Proposal F advocates a cautious stance, urging contributors to refrain from using generative AI unless absolutely necessary.

Proposal G aims to guarantee that direct contributions to Debian are human‑authored—meaning code should be written manually—while permitting the use of LLM assistance in other aspects of the development process. It concludes by disallowing direct output from generative AI as contributions.

Finally, Proposal H zeroes in on a single but critical ethical concern: the environmental impact. It states that LLM usage accelerates the destruction of our ecosystem and therefore constitutes a deal‑breaker.

If Proposal H has a weakness, it is that it does not differentiate between local and cloud‑based LLMs, although one could argue that the environmental cost of training is the dominant factor, and whether the final model runs locally (relatively efficient) or remotely (highly inefficient) is secondary.

From a neutral perspective, the sheer number of proposals, many overlapping, may hinder the likelihood of achieving a 3:1 supermajority for the first option. A more streamlined set of distinct positions would likely facilitate a clearer decision. Nevertheless, this is an internal matter for Debian contributors.

Voting remains open for a few more days; the tally is steadily rising, with fewer than 350 ballots cast at the time of this report.

Several other projects have already taken a stance. In April 2024, Gentoo prohibited AI‑generated code, a position later adopted by NetBSD. The OpenBSD project, as noted in our coverage of its latest release, has determined that AI‑generated code cannot be copyrighted and therefore cannot be merged into the repository; however, it has grandfathered code from the Tmux project, which permits LLM‑assisted contributions, making the restriction somewhat flexible. Additional projects may follow suit.

In September 2025, we reported that FreeBSD had expressed opposition to AI‑generated code, though it has yet to formalize an official policy. Recent social media discussions suggest an ongoing internal debate, potentially involving a vote akin to Debian’s, though details remain private.

While some distributions may impose bans on LLM‑generated code, the practical impact remains uncertain. The Linux kernel has officially distanced itself from an anti‑AI stance, and Red Hat, an IBM subsidiary that contributes a substantial portion of code to many distributions, has fully embraced AI tools. Consequently, the decisions of individual distros may have limited overall effect. ®

Source link