Debian Puts LLM Contributions to a Vote, With One Proposal Calling for a Ban

The Debian project is holding a General Resolution to decide how, or whether, large language models can be used in its work, with competing proposals (A through D) on the ballot. The document lays out Proposal A in full: an outright prohibition on any contribution written or assisted by generative AI. Its scope covers Debian source packages, native project software like lintian, official web resources, and contributor-written documentation and translations, while explicitly exempting upstream projects, AI-related software, and upstream patches. The measure would go so far as to add a new clause to Debian’s Social Contract formalizing the ban.

Proposal A rests on four arguments. Copyright is unsettled — LLM output has murky licensing that clashes with Debian’s requirement for absolute clarity under its Policy and the DFSG. Quality is unreliable, since a model produces statistically likely text rather than verified fact, which is especially dangerous in packaging where each package is unique and generated output tends to blend incompatible conventions across the archive’s history. Community health suffers when new contributors submit AI output that reviewers must clean up, burning out maintainers without the newcomers actually learning the craft. And ethically, the proposers point to AI companies scraping the web without regard for licenses or robots.txt, an activity they say has functioned as a sustained denial-of-service against Debian’s infrastructure and forced defensive measures like JS-based bot checks.

The proposers concede enforcement would be difficult, framing the ban as a good-faith statement of intent rather than a technically policed rule. The move follows similar decisions at GNOME, Gentoo, and Codeberg, signaling that generative AI has become a governance flashpoint across free-software communities weighing tooling convenience against provenance, quality, and the strain on volunteer reviewers.