Debian developers vote on whether to ban or permit AI-assisted contributions under new rules

Debian’s general resolution introduces eight competing proposals to address AI-assisted contributions, ranging from an outright ban to conditional acceptance. The strictest option, Proposal A, would amend the Debian Social Contract to prohibit LLM-generated contributions in packaging, documentation, and project communications. This reflects concerns about copyright ambiguity, reliability, and the additional review burden on maintainers. However, it exempts upstream projects, meaning AI-generated code could still enter Debian indirectly through dependencies.

Permissive proposals, such as B and D, allow AI-assisted contributions if they meet quality, security, and licensing standards, with mandatory disclosure of AI involvement. These options acknowledge enforcement challenges and the impracticality of policing tool usage. Proposal E takes a neutral stance, applying the same contribution standards regardless of origin. The debate highlights a tension between preserving human oversight and accommodating tools that may improve efficiency but introduce risks.

A unique proposal, labeled 'Debian is created by humans,' shifts focus from banning tools to restricting AI-generated output in official work. Contributors could use AI for research or analysis but not submit its output directly. This approach avoids outright prohibition while maintaining human accountability. Another strict option, Proposal H, would require all Debian communications to be human-written, with AI use disclosed and subject to Code of Conduct enforcement. The diversity of proposals underscores the lack of consensus on how to integrate AI without compromising project integrity.

The voting thresholds differ: Proposal A requires a 3:1 supermajority, while the others need only a simple majority. This suggests the project views a ban as a more consequential change than permissive or neutral rules. External commentary frames the debate as a symptom of broader uncertainty about verifying AI-generated work, particularly in critical infrastructure. The outcome may influence other open-source projects grappling with similar questions, as AI tooling becomes harder to detect and regulate.