Debate on the Use of LLMs in Debian: Ban or Conditional Approval?

The Debian Project is intensifying its debate over whether the use of large language models (LLMs) should be permitted in code development. A voting page published on July 25, 2026, presents two contrasting proposals, sparking varied opinions within the community. One proposal advocates for a strict ban on the use of LLMs within Debian itself, while the other supports conditional approval that allows LLM-generated contributions under specific compliance and responsibility guidelines.

Two Opposing Proposals

The first proposal (Proposal A), submitted to the Delegates’ voting page (Debian Vote 2026-002), seeks to completely ban the use of LLMs within the scope of the Debian Project. However, this ban is limited to Debian itself and does not extend to upstream projects or upstream security patches. In other words, Debian maintainers would be prohibited from using LLMs to write package code directly, but Debian could still accept code written using LLMs by upstream developers.

The opposing proposal (Proposal B) does not aim to prohibit AI-assisted contributions outright but instead advocates for managing them within a framework of quality assurance and compliance. Specifically, it sets forth three requirements:

1. Verification that AI-generated code complies with Debian’s various guidelines.
2. Individual contributors must assume responsibility for the code.
3. A requirement for disclosure when contributing AI-generated code.

Developers’ Concerns

Debian developers have expressed multiple specific concerns regarding the use of LLMs.

First, unresolved copyright and licensing issues surrounding LLM-generated code are a major concern. It remains unclear how the licenses of the code included in the training data for LLMs might affect the generated output. This lack of clarity raises questions about the compatibility of such code with free software licenses like the GPL.

Second, ensuring the quality and accuracy of LLM-generated code is challenging. While LLMs produce statistically plausible outputs, these outputs do not always reflect specialized knowledge or intent accurately, leading to potential bugs and security vulnerabilities.

Third, there is concern about the increased burden placed on community reviewers. If LLMs generate large volumes of code, it may become impractical for human reviewers to meticulously scrutinize all contributions. This issue is particularly critical for large-scale distributions like Debian, which maintains over 59,000 packages and already faces strained review resources.

Fourth, there is criticism of AI companies for disregarding copyright and licensing requirements when scraping data to train large models, which is seen as harmful to the free software community as a whole. Developers have also highlighted issues such as these companies conducting what amounts to distributed denial-of-service (DDoS) attacks on websites heavily relied upon by users and the enormous energy consumption involved in training such models.

Impact on the Free Software Community

The relationship between LLMs and open source has grown increasingly tense in recent years. Notable examples include the class-action lawsuit against GitHub Copilot and Stack Overflow’s temporary ban on AI-generated code submissions. This vote by Debian represents an attempt to establish a clear policy at the distribution level against this backdrop.

A similar precedent exists in the Linux kernel community, which strengthened its rules regarding AI-related contributions in 2025. As demonstrated by the Debian 13.6 release, Debian, known for its focus on stability and reliability, has consistently exercised caution when adopting new technologies.

Editorial Opinion

In the short term, the outcome of this vote will directly affect the development cycle of Debian 14. If the ban proposal is approved, Debian maintainers will need to revert to writing code entirely by hand. On the other hand, if the conditional approval proposal passes, the community will need to quickly establish mechanisms for disclosure obligations and quality reviews. In any case, further debates about the effectiveness of the policy are inevitable in late 2026 and early 2027.

From a long-term perspective, this discussion is likely to serve as a precedent for the formation of AI usage norms across the open source community. Debian’s decision could influence derived distributions like Ubuntu and other projects, and may eventually impact the guidelines of organizations like the FSF and OSI. The crux of the matter lies in determining whether AI should be treated as a collaborator or merely as a tool—a question that could drive changes in copyright law and licensing frameworks.

As an editorial note, it is essential to address the practical issue of who will conduct code reviews for LLM-generated content and based on what standards, as this remains an unresolved challenge.

References

• “Debian 讨论是否允许使用 LLM”, by (author unknown) — Solidot, 2026-07-25T15:53:13.000Z (ARR)
• Source URL: https://www.solidot.org/story?sid=84922