2 min read
In short
A thousand Debian developers are deciding whether LLMs can be used when working on the distribution. But the line has been drawn such that AI-generated upstream code is allowed, while original work is in question. I’m exploring why this matters for the entire open-source ecosystem.
Debian has launched a general vote among developers on the use of AI tools and large language models in the development of the distribution. About a thousand people are voting—those who maintain packages and support the project’s infrastructure. There are two options on the table: a complete ban or permission with conditions.
The most interesting aspect here isn’t the vote itself, but where Debian has drawn the line. The vote applies only to internal work: packages developed within Debian, web resources, translations, documentation, and official correspondence. At the same time, upstream projects that use AI remain outside the scope of this discussion. AI-related packages will continue to be distributed. Patches created using AI in upstream projects are also accepted.
This creates an interesting situation. If a developer of an upstream project generates part of the code using an LLM and submits a patch, Debian will accept it. But if a Debian package maintainer uses AI themselves to write a patch or documentation—that may be prohibited. The line isn’t drawn based on code quality, but rather on at what point in the workflow the AI was applied.
This isn’t just a whim on Debian’s part. Behind this issue lies a real problem: license compliance, copyright, and trust in the code. LLMs are trained on corpora whose origins aren’t always transparent, and a project of Debian’s caliber cannot afford a legal gray area in its base packages. A complete ban is a radical but legally safe response.
The other side of the coin is productivity. There aren’t many Debian maintainers, but there are tens of thousands of packages. A ban on AI tools means that people will have to do manually what their colleagues in commercial projects have already automated. The question is how sustainable this is in the long term.
The vote is not yet over, and it is unclear which option will prevail. But the very fact that one of the oldest and most influential open-source projects is formally defining its stance on AI will set a precedent. Other distributions and communities will be watching the arguments from both sides.
For developers working with open source, this is a signal: the era when the use of AI in code was a personal matter for each individual is coming to an end. Projects are beginning to formulate rules, and these rules will vary.