QGIS closed the gate.
QGIS merged version 2.0 of its AI tool policy on October 7. The policy says that no LLM, AI, or agent-assisted code contributions are allowed in the QGIS repository. It also warns that later discovery can lead to code reversion and contributor sanctions.[1]
This is a project rule, not a scientific finding about code quality. The accepted proposal replaced a longer human-in-the-loop policy with seven added lines and 140 deleted lines. The pull request author called it a strict "NOT ALLOWED" policy. That change is narrow, blunt, and easy to cite during intake.
A contribution can be cheap to produce and expensive to understand. The unpaid side of that trade gets to set the gate.
The reason lives in the review queue.
In the proposal discussion, QGIS contributors described more AI-assisted submissions than maintainers could review with care. They also argued that a contributor who cannot explain and revise a change does not grow into a future reviewer. One contributor who had submitted AI-assisted fixes agreed with the ban and wrote that the tools had given them an illusion of readiness for a complex codebase.[2]
The earlier policy tried disclosure, self-review, accountability, and an "extractive" label. Participants in the replacement discussion said those steps did not work in practice. The new policy starts with a closed door. Its author said later proposals could argue for specific exceptions.
The door is already under debate.
QEP 429 is an open proposal, not current policy. It would allow AI-assisted contributions from people with either project voting rights or at least 100 accepted pull requests. The author describes that as a way to keep random submissions out while letting experienced contributors choose their tools.[3]
Objections in that thread cover review cost, copyright, unequal treatment, environmental cost, and the role of a human community in free software. Supporters argue that known contributors have earned trust through prior work. No one should report the exception as adopted while the pull request remains open.
A moratorium needs a reopening test.
A pause can protect a tired review team. It can also harden into an unwritten permanent rule if the project never names the evidence that would change it. Write down the scope, enforcement route, review budget, and next decision date. If exceptions are possible, state who can propose them and what a trial must measure.
Do not use an AI detector as the enforcement plan. QGIS frames compliance around contributor conduct and maintainer judgment. A useful intake rule should say what the contributor must own, what the reviewer may reject without investigation, and what happens after a violation.
Steal this policy checklist.
- Name the exact repository and contribution types covered.
- State whether the rule is a pause, a permanent ban, or an evidence-gated trial.
- Give maintainers a short rejection path that does not require origin forensics.
- Define contributor duties in observable terms: explain, reproduce, revise, and maintain.
- Record review capacity before opening an exception.
- Set a review date and the evidence needed to loosen or tighten the rule.
The brake rack below drafts those fields. It does not judge QGIS, detect AI use, verify authorship, or create policy for another project.