Debian developers vote on banning LLM-assisted contributions

Debian developers are voting on a General Resolution to ban LLM-assisted contributions to the project. Developer Matthias Geiger, with Jesse Rhodes, filed the original proposal, which would bar generative AI from Debian source packages, packaging work, documentation, translations, and project communications. Patches and security fixes pulled in from upstream projects, and upstream projects that use AI themselves, would be exempt.

The resolution's stated reasoning covers four points: unresolved copyright status for LLM output, quality and accuracy risk, reviewer burnout from a rising volume of AI-generated submissions, and objections to models trained on web content without license compliance.

A ban is only one of five competing resolutions on the ballot. The alternative getting the most attention would keep AI-assisted contributions allowed but require the contributor to take full responsibility for quality, security, and licensing, be able to explain everything they submit, and flag AI involvement in commit metadata, disclosure and accountability instead of prohibition.

For a maintainer running a smaller open source project, the specifics of Debian's ballot matter less than the shape of the decision: whichever resolution wins becomes a template other projects will borrow from, the same way Debian's packaging and governance conventions already get reused well beyond Debian itself. A project without a stated AI-contribution policy today is a maintainer without an answer the next time a pull request shows up with an unstated LLM behind it.