Debian Deliberates AI Contributions Without Reaching Decision

In March 2026, the Debian project found itself at the center of a growing conversation within the free software community about how to handle contributions generated by artificial intelligence. After Lucas Nussbaum opened a discussion with a draft general resolution (GR) on whether Debian should accept AI-assisted contributions, the project engaged in what turned out to be an illuminating but ultimately inconclusive debate.

The proposed GR would have allowed "AI-assisted contributions (partially or fully generated by an LLM)" under specific conditions. These included requiring explicit disclosure when "a significant portion of the contribution is taken from a tool without manual modification" and labeling such contributions with "a clear disclaimer or a machine-readable tag like '[AI-Generated]'." The resolution also would have required contributors to "fully understand" their submissions and be accountable for "including vouching for the technical merit, security, license compliance, and utility of their submissions." Additionally, it would have prohibited using generative-AI tools with non-public or sensitive project information.

One of the earliest challenges in the discussion was pinning down terminology. As Russ Allbery pointed out, "AI is the catch-all term, but much (not all) of the technology in question is actually tooling around large language models (LLMs)." He argued that "AI just means whatever the person writing a given message wants it to mean and often changes meaning from one message to the next, which makes it not useful for writing any sort of durable policy."

This terminology debate highlighted a fundamental challenge: how can a project create policy around technology when participants can't agree on what to call it? Sean Whitton suggested narrowing the focus to "LLM" rather than the broader "AI" and distinguishing between different uses such as code review, generating prototypes, or generating production code.

Beyond terminology, the discussion revealed deeper concerns about the impact of AI tools on the Debian community. Simon Richter raised what he called the "onboarding problem": "An AI agent, he said, could take the place of a junior developer. Both could perform basic tasks under guidance, but the AI agent would not learn anything from the exchange; the project resources spent in guiding such a tool do not result in long-lasting knowledge transfer."

Richter argued that accepting AI-assisted "drive-by contributions" is harmful because it misses opportunities to onboard new contributors. "The best-case outcome is that a trivial problem got solved without actually onboarding a new contributor, and the worst-case outcome is that the new contributor is just proxying between an AI and the maintainer."

Matthew Vernon introduced ethical concerns about the companies developing AI tools. "The organizations that are developing and marketing tools like ChatGPT and Claude are behaving unethically, he said, by systematically damaging the wider commons in the form of automated scraping and doing as they like with others' intellectual property. 'They hoover up content as hard as they possibly can, with scant if any regard to its copyright or licensing'."

The discussion also touched on copyright questions, both regarding the licenses of material used to train models and the copyright status of LLM outputs. Jonathan Dowland suggested it might be better to forbid some contributions now when the legal situation is unclear, while Thorsten Glaser took a particularly harsh stance against LLM-driven contributions, suggesting some upstream projects should be moved out of Debian's main archive.

A technical point of contention was the question of "preferred form of modification" for code generated through AI tools. Bdale Garbee asked, "what is the preferred form of modification for code written by issuing chat prompts?" Nussbaum responded that this would be "the input to the tool, not the generated source code." However, this raised questions about reproducibility, given that LLM outputs can vary even with the same input due to their non-deterministic nature.

The conversation ultimately did not result in a formal GR being put forward. On March 3, Nussbaum indicated that while he had initially proposed the GR "in response to various attacks against people using AI in the context of Debian," the discussion had been civil and productive enough to continue on mailing lists rather than rushing to a vote. "As long as the discussions around AI remained calm and productive, the project could just continue exploring the topic in mailing-list discussions."

The diversity of opinions within the Debian community reflects broader uncertainties across the free software ecosystem about how to handle AI-generated content. While some see potential benefits in using these tools to accelerate development and make complex tasks more accessible, others raise valid concerns about quality, ethics, copyright, and the impact on community development.

For now, Debian will continue handling AI-generated contributions on a case-by-case basis using existing policies. This pragmatic approach may be the most sensible given the rapidly evolving nature of AI technology and the lack of clear legal precedents. As Nussbaum noted, "if there were a GR, 'the winning option would probably be very nuanced, allowing AI but with a set of safeguards'."

The questions raised in this debate—about terminology, quality control, ethical implications, copyright, and community impact—are likely to remain relevant as AI tools become more prevalent in software development. Debian's measured approach provides a thoughtful example for other projects grappling with these same issues.

For those interested in following this ongoing discussion, the Debian mailing list archives contain the full conversation. The Debian Wiki also maintains pages on various policy discussions that may be updated as this conversation continues.