QGIS, the open-source geographic information system, has changed its AI policy to "not permitted." The proposal, numbered QEP 428, was marked Accepted and merged into the QGIS Enhancement Proposals repository on October 7. Its title is a single sentence: No AI contributions permitted.
The proposal was opened by QGIS core developer Nyall Dawson on September 24 and went to a vote on October 5. In the voting thread he lists four members who voted in favor, and several others explicitly gave a +1 during the discussion; the opening post drew 25 thumbs-up and 3 thumbs-down, and about 15 people took part in the discussion overall. QGIS is one of the most widely used open-source desktop GIS applications in the world, and after this vote it no longer accepts code contributions written by AI.
From "human in the loop" to a blanket ban
QGIS had been operating under QEP 408, whose core principle was "human in the loop": AI tools were allowed as long as a person took responsibility for the submission. Dawson had already said on the developer mailing list in June that these rules were "rather permissive," and he pointed to the conflict the PostGIS community had run into over AI submissions at the time.
The new proposal gives two main reasons. The first is that maintainers cannot keep up: the volume of AI-generated submissions has outgrown what people can review, so careful review is impossible, and waving things through harms code quality. The second is that an open-source community is held together by relationships between people, and room needs to be kept for human contributors.
How the old policy fared in practice also came up. An "extractive" label that had been planned never got created. Member uclaros said in the discussion that around 50 PRs had been tagged "AI Assisted," and roughly 40 of them were ultimately closed without being merged. That figure comes from the discussion thread; the page offers no item-by-item list, so it cannot be independently verified for now.
Dawson describes the new rule as closing the door first and then opening it again case by case later, for example using AI in place of IDE autocomplete or for translating interface text, provided it is discussed thoroughly beforehand.
How it compares with GDAL and Godot
Within the same circle, GDAL got there first. Its rules state that "Submission of vibe-coded contributions is banned." A human must be the primary author of a contribution; large language models may serve only as enhanced autocomplete, or for mechanical refactoring that could otherwise be done with deterministic algorithms; unattended agent submissions are never accepted; and any use of an LLM must be disclosed in the ticket, a comment or the commit message.
The game engine Godot leans toward accountability. Its core line is "Contribute only what you understand": submitters must be able to explain their code to maintainers, and whether they used AI, copied someone else's code or implemented someone else's idea, they must say which parts they did not write themselves.
When Dawson posted in June, he suggested a combination modeled on GDAL plus Godot. The QEP 428 that was finally adopted is stricter than both: GDAL at least leaves room for autocomplete and mechanical refactoring, whereas this version of the QGIS rule defers even those to a separate discussion later on.
How it will be enforced is still unclear
On enforcement, there are only principles so far. Member timlinux said in the discussion that the standard is whether you wrote the code yourself in good conscience, and that the project will not inspect anyone's computer. In other words, the rule depends mainly on contributors' self-discipline, plus the judgment of maintainers during review.
Several gaps remain at the edges. vpicavet and adamjstewart each asked whether the rule also applies to non-code content such as issues, comments, PR descriptions and documentation, and the discussion was also divided on where to draw the line between "using AI to write code" and "using AI to help with troubleshooting or review." As of the proposal's merge, no formal answer was visible on the page.
For developers outside the project who contribute to open source, the direct effect of the new rule is this: before opening a PR against the main QGIS repository, code generated by tools such as Copilot or Cursor can no longer simply be carried over. Many GIS, surveying and government GIS projects in China are built on top of QGIS; how they use AI in their own repositories is not restricted, but patches contributed back upstream will have to follow the new rule.
Sources: QGIS Enhancement Proposals repository (QEP 428 discussion page), QGIS developer mailing list, CocoLoop, GDAL project AI policy; verified the proposal number, the number of supporting votes, and the basis for the figure of about 50 AI Assisted PRs.