Debian will allow generative AI if humans stay accountable
A general resolution chose "Responsible Use of Generative AI." Disclosure is encouraged, not required.
Debian has voted. The results were posted on 29 August. LWN carried them. The winner was choice 5, titled Responsible Use of Generative AI.
The project neither endorses nor prohibits generative AI. Contributions still have to meet the same standards of quality, correctness, maintainability, and legal compliance.
The contributor stays responsible. They must understand, review, test, and modify the output. Blind acceptance is inconsistent with Debian's practices.
Disclosure of AI assistance is encouraged. It is not required.
Contributors must not share confidential or security-sensitive information with third-party AI.
Options for a full ban failed to beat None of the Above.
Debian is one of the large free-software projects that other systems still rest on. A rule here is a signal, even when it is a middle path.
The vote did not bless the tools, and it did not throw them out. The human who uploads the work stays on the hook.
The policy is simple. Generative AI is neither endorsed nor prohibited. AI output is not a special class of contribution. Full ban options failed to beat None of the Above.
Encouraged disclosure without a requirement means reviewers may not know how a patch was made. The standard they apply is supposed to be the same in any case. Contributors must understand the work, review it, test it, and modify it. Blind acceptance is inconsistent with Debian's practices.
The confidentiality rule is the other hard edge. Security-sensitive material and other confidential information stay off third-party AI. That limit is on the workflow.
The vote is done. Contributors who use the tools stay accountable for the work. Confidential material stays off third-party AI. That is the rule from here.