QGIS bans every AI-assisted code contribution to its main repository

A GIS developer who runs Claude Code or Codex over a QGIS patch and opens a pull request can now be reverted and lose contributor rights for doing it. QGIS merged a new version of its AI tool policy, QEP-408 version 2.0, into its enhancement proposals repository on 7 October, after a four-week discussion on the OSGeo qgis-developer mailing list.

The policy, written by Nyall Dawson and signed by the QGIS project contributors, replaces the previous version by Even Rouault from earlier in the year, which had asked contributors to label AI-assisted work but did not forbid it. The new text is three sentences. It says: "No LLM, AI or agent assisted code contributions are allowed in the https://github.com/qgis/qgis repository. If we subsequently discover that you have used an AI, LLM, or any agent-based workflow for your contribution, you will be subject to code reversion and contributor rights sanctioning."

How it got there

The discussion opened with a proposal from Dawson to the qgis-developer list: a complete moratorium on LLM-based contributions and a "proof of life" requirement for new contributors, after what he described as considerable hurt in the community over an earlier incident. Tim Sutton, a QGIS project steering committee member and the chief executive of QGIS consultancy Kartoza, replied that the PSC was fine with a ban if that was what developers wanted, and offered to help write the enhancement proposal.

Dawson's reasoning on the list was that the project should start from a safe no-AI default and later loosen it, rather than keep a loose policy and try to tighten it. The merged document itself gives no rationale, which Sutton had asked for: he wrote on the list that putting a reason in the text would invite arguments about nuance, and that the "what" is enough.

What a reader has to do

A contributor who has used an AI assistant to write or edit code they planned to submit to the main QGIS repository should not open the pull request. The policy names LLMs, AI, and any agent-based workflow, which covers everything from editor auto-complete to a background agent writing the whole patch. Enforcement will rely on reviewer judgement and on contributors telling the truth, with reversion and sanctions as the response when AI use is found later.

The policy applies to github.com/qgis/qgis. It does not list other QGIS repositories, so a plugin, a translation or a documentation change in a different repo is not covered by this text. A team planning to automate contributions through an agent should assume the rule will extend if the current ban holds.

A team that submits patches to several open-source projects now has to read each one's contributor policy before running an agent over the diff. The policy is three sentences on purpose, with no exemption for trivial fixes, no label workflow, and no size threshold. A reviewer who finds AI use after a patch is merged has one response in the policy, which is reversion.

Source

The merged policy file is qep-408-ai-tool-policy.md on master. The pull request is #387 by nyalldawson. The thread that produced it is on the OSGeo qgis-developer mailing list, with Tim Sutton's reply as PSC member.