Debian’s August vote selected option 5, Responsible Use of Generative AI. The adopted text keeps responsibility with contributors and encourages disclosure of AI assistance. It does not make generated output automatically acceptable.
Read the winning text, not all nine options together
The official result and resolution identify option 5 as the winner. It neither endorses nor prohibits generative AI use. Contributors remain responsible for understanding, reviewing and testing submitted work. Disclosure is encouraged, not mandatory. Existing quality and licensing expectations still apply; sensitive information and broad automated actions also require appropriate care and project process.
The unsuccessful alternatives on the same page are proposals, not additional adopted rules. Combining sentences from them would create a policy that Debian did not vote for.
The result is a ranked comparison
The published matrix reports 203 preferences for option 5 over option 2, against 148 in the opposite direction. These are pairwise preferences. They are not first-choice vote totals and cannot be added to another matchup to count distinct voters.
Consider a tiny illustrative ballot with three people ranking A above B above C. All three prefer A to B and A to C. Adding those two counts gives six preferences but still only three people. This is why an election matrix must be read by matchup.
For the real election, the secretary's statistics list 425 unique voters. That denominator is a different quantity from the number expressing a strict preference in any one comparison.
What a useful assisted contribution looks like
Suppose a generated patch changes how an empty configuration value is handled. A useful submission explains the previous behavior, the intended behavior and which cases were checked. It should identify why an empty value differs from a missing value, rather than merely attaching a large diff and saying that a model found a bug.
A reviewer can then inspect an actual claim. The author should be able to explain each meaningful change and remove unrelated edits. If the patch does not address a reproducible issue, generating more variations will not supply the missing evidence.
These are practical submission examples, not a new Debian checklist or a mandatory AI label invented by this article.
What the decision does not promise
The resolution does not grant an automatic right to merge a contribution. The earlier version of this article went too far by saying a reviewer could no longer reject work because a tool was involved. The adopted text does not establish that guarantee.
When contributing, read the relevant team's current instructions and discuss a large proposed sweep before producing hundreds of patches. The question is whether maintainers can evaluate and maintain the work being offered. An AI tool can help its author; it cannot take over that author's responsibility to the project.
Correct unsupported claims that reviewers must accept AI work; explain the adopted option and pairwise vote counts.