Debian will not ban the use of generative AI for writing code, documentation, and other contributions to the open-source project. Developers may use AI tools as long as they can vouch for the results themselves. Disclosing the use of AI is encouraged but not required.
This settles, for now, a discussion within the Debian community about how to handle generative AI, according to The Register. In a vote, developers were able to rank eight different proposals. These ranged from a ban to variants in which AI was permitted under certain conditions. Nearly 450 votes were ultimately deemed valid.
The proposal “Responsible Use of Generative AI” received the most support. Its core principle is that Debian does not want to distinguish between work created entirely by a developer and contributions made using AI. The same requirements apply to both, including those related to quality, accuracy, maintainability, and legal aspects.
Developers remain responsible
This does not mean that generated code can be added without further ado. Anyone using AI must understand and verify the output before it is submitted to Debian. Testing and, where necessary, making adjustments are also explicitly the responsibility of the developer.
Debian thus rejects the argument that errors can be attributed to an AI tool. According to the adopted proposal, accepting generated code without human review is not in line with the project’s development practices.
At the same time, Debian does see benefits in the technology. AI tools can take work off volunteers’ hands, thereby freeing up time for tasks where technical knowledge, judgment, and collaboration are more important. The project therefore opts not to impose restrictions based on the tools used, but rather to set requirements for the final result.
A notable aspect of the policy concerns transparency. Debian encourages developers to disclose when AI was used in a contribution. However, such disclosure is not mandatory. The project thus relies largely on the existing sense of responsibility among individual contributors and the review process.
Open-source projects take different approaches
Within the open-source world, there is still no common approach to AI-generated code. Gentoo has banned the use of generative AI for contributions. NetBSD and OpenBSD also take a much more restrictive stance toward AI-generated code.
The Linux kernel, on the other hand, takes a more open approach. Linus Torvalds has previously made it clear that Linux does not consider itself an anti-AI project. He himself recently used AI to solve a difficult software problem.
That doesn’t mean Torvalds applauds every use of AI. Earlier this year, he complained about the volume of automatically generated bug reports on the Linux security mailing list. According to him, these made it increasingly difficult to distinguish useful reports from noise.
Debian’s decision thus amounts to a middle ground: the technology used to create a contribution is not the deciding factor, but human oversight remains so. AI may assist with the work, but it does not alter the responsibility of the person who ultimately submits the code.