Debian votes to allow "responsible use of generative AI"

TL;DR Highlight

Debian voted to neither ban nor endorse generative AI, keeping contributors fully responsible for submissions.

Who Should Read

A developer who submits AI-assisted code to open source projects or must set an AI code policy for their team.

Core Mechanics

Option 5 won the vote
- In Debian's general resolution, 'Responsible Use of Generative AI' came first.
- The article was posted on August 29, 2026.
Debian neither endorses nor bans AI
- It takes no side on generative AI tools in development, maintenance, documentation and packaging.
- It acknowledges they can substantially improve contributor productivity when used responsibly.
Same standards regardless of tools
- Quality, correctness, maintainability and legal compliance standards apply equally.
- Using an AI tool does not reduce the submitter's responsibility.
Contributors must verify AI output
- Contributors are expected to understand, review and test AI-assisted output.
- They should modify it where appropriate before submitting to Debian.

Community Reactions

Opinion that the policy boils down to 'your code, your responsibility'
- Several commenters agreed that submitted code is the author's responsibility, AI or not.
- One suggested treating generated code as if you typed every character yourself.
Explanation that hardline anti-AI options lost even to 'None of the Above'
- A commenter said options 1 (social contract change) and 3 (code of conduct change) ranked below NOTA.
- Another countered that option 3 allowed AI contributions while discouraging them, and aimed to signal Debian's stance.
Experience that local coding models are now usable
- One commenter said MIT-licensed small and mid-sized models match Opus 4.5, and Qwen3.8 27B mostly fits in 32GB VRAM.
- They noted the largest commercial models are still out of reach and a knowledgeable developer should stay in the loop.
Opinion that self-assessed AI level on contributions is useful
- It lets the receiver decide how much effort to spend understanding the code.
- A VisiData blog post was linked.

How to Apply

Add an AI-use checkbox to your PR template
- Today, add a field for AI usage level and what you tested yourself.
- Let reviewers decide how closely to look.
Write your AI code policy as one principle
- Put 'the submitter is responsible for understanding, reviewing and testing, regardless of tool' in CONTRIBUTING.
- Prefer an enforceable accountability rule over a ban.
Reject AI code the author can't explain
- During review, ask the author to explain why each change was made.
- Require a revision if they can't.

Try It

copy & run

# Paste into Claude Code or ChatGPT (a simplified version of Debian's principle as a PR self-review)

You are a strict reviewer. The diff below was written with AI assistance. Check whether the submitter can claim to have understood, reviewed and tested it.

{{diff}}

Answer:
1. Parts that may be problematic for quality, correctness, maintainability or licensing
2. Three key changes the submitter must be able to explain personally
3. Additional tests needed

Next step: answer the questions yourself and re-read any part you struggle to explain.

Terminology

general resolutionA procedure where all Debian developers vote to set project policy. Like a shareholder meeting vote.

None of the AboveA ballot option meaning 'none of these are acceptable'. Options ranked below it are effectively rejected.

Condorcet winnerAn option that beats every other option head-to-head. Like a team that wins every match in a tournament.