Open-Source GIS Project QGIS Adopts a New Policy: No More LLM-, AI- or Agent-Assisted Code Contributions, Citing Review Burden and Community Trust

On October 7 the open-source geographic information system QGIS merged enhancement proposal QEP 428, replacing the AI policy from QEP 408 with a strict "not allowed" rule: according to a related project commit, the qgis/qgis code repository no longer accepts LLM-, AI- or agent-assisted code contributions. Core contributor Nyall Dawson submitted the proposal on September 24 after a mailing-list discussion; it moved to a vote on October 5 and was merged after approval from several maintainers, with the proposal post drawing 86 thumbs-up and 24 thumbs-down reactions. Supporters cited a growing review burden on maintainers, declining review quality and protecting trust between community members and contributors; Dawson said the old policy wasn't working and the project had "already lost at least one extremely high value contributor." The policy has no end date and relies on contributors' self-attestation; whether it covers issues, docs and translations isn't settled, and another proposal under discussion would extend it to all text in the repository.

The policy

QGIS is one of the most widely used open-source geographic information systems. On October 7 the project merged enhancement proposal QEP 428, titled simply "No AI contributions permitted." It replaces the AI policy from QEP 408 with a strict "not allowed." In the wording of a related project commit, the qgis/qgis code repository no longer accepts code contributions written with LLM, AI or agent assistance, and violations can lead to code being reverted and contributor rights being restricted.

Long-time core contributor Nyall Dawson submitted the proposal on September 24 after a mailing-list discussion; it went to a vote on October 5 and was merged after approval from several maintainers. The proposal post drew 86 thumbs-up and 24 thumbs-down reactions.

Why

Supporters' reasons fall into two groups: review burden, as AI-generated submissions increase maintainers' review load and reduce review quality; and community trust, since open-source projects run on trust between contributors. Dawson put it most directly: the old policy wasn't working, and "we've already lost at least one extremely high value contributor."

What isn't settled

The policy has no end date. Maintainers in the discussion called it an initial policy, and Dawson described it as closing the door first, with room to open it selectively later, for example for AI as an autocomplete replacement or for translations. Enforcement can only rely on self-attestation: one maintainer said plainly that the test is whether you wrote the code yourself, and the project won't inspect contributors' machines.

Scope is also contested. Commenters asked whether issues, comments, docs and PR descriptions are covered, and some argued the wording is so broad it could ban legitimate tools; another proposal still under discussion would extend the policy to all text in the repository.

This is the same pressure as Google pausing its open-source bug bounty, covered here on October 5: AI makes submitting nearly free, while the cost of checking falls on maintainers. For QGIS contributors, the safest course for now is to write the code yourself, and to ask the community first if you're unsure whether a tool counts as AI assistance.

via: QEP 428 proposal (GitHub)