QGIS proposal to ban AI-assisted contributions draws pushback over broad wording

QGIS, one of the most widely used open-source geographic information systems, is weighing a formal ban on contributions produced with AI help. The change has been filed as QEP428 in the project's enhancement-proposals repository by contributor nyalldawson, and after surfacing on the Hacker News front page it quickly opened a debate about where the line should sit.

What the proposal would do

According to the GitHub pull request that Hacker News surfaced, the change adds a new entry to the QGIS Enhancement Proposals, numbered 428 and titled "No AI contributions permitted." The full policy text lives in the pull request, but the surrounding discussion shows it hinges on broad language about contributions that are AI-"assisted" or in which AI is "used" — terms that one commenter flags as expansive enough to cover almost any contact with a language model.

The objection: losing human-in-the-loop tooling

The most detailed pushback comes from a community member commenting on the pull request, and their stance is notably not pro-AI triumphalism. They see the current frenzy around AI as deeply damaging in ways that go well beyond code, and they acknowledge the strain its misuse puts on maintainers and developers. Their complaint is narrowly technical: as written, the policy appears to prohibit uses of language models in which a human remains fully in control.

They read the ban as extending to review, auditing and security work. Drawing on their own tentative steps into QGIS's C++ core, they describe two workflows they would lose. One is using models to uncover context in a large codebase that is easy to miss — the output is frequently incorrect, but it helps a human build understanding that feeds into better contributions. The other is treating a model as a stand-in reviewer: paste code in, receive a list of findings, discard the irrelevant ones, fix the genuine issues, and open a stronger pull request as a result. They compare an in-editor model to a mildly useful assistant offering occasional feedback that cuts down errors, and liken its reliability to web search, which mixes accurate and inaccurate results and leaves verification to the person reading them.

The commenter also presses on intent. Whether the policy is meant as a political or ethical statement, as relief for overworked developers, or as a general stand against AI, they argue it should say so explicitly — and that whichever goal applies, the current wording could be both clearer and stronger.

The governance problem underneath

The exchange points at the difficulty now facing maintainers of large open-source projects. Almost all modern development is tool-assisted to some degree — autocomplete, search, documentation lookups — so a rule defined by "assistance" needs a precise boundary before it can be enforced consistently. Draw it too loosely and reviewers keep wading through machine-generated pull requests that look plausible but do not work; draw it too tightly and contributors who use models as study aids or pre-submission checkers, while fully understanding the code they ship, have little reason to stay. The commenter's core concern is that the policy as drafted would exclude functionally sound, human-reviewed contributions simply because a model was involved somewhere in the process.

Why it matters

If adopted, QEP428 would make QGIS one of the most prominent open-source projects to formalize a blanket restriction on AI-assisted contributions, and other projects would likely study the wording closely. The argument it has triggered previews a conversation every sizable codebase will need to have: what actually counts as an AI contribution, who is responsible for verifying it, and whether a ban protects maintainers or mainly pushes careful contributors toward silence or the exit. Whether QGIS keeps the broad language or carves out human-in-the-loop uses such as review and code exploration will be a data point for the wider ecosystem.