QGIS, the popular open-source geographic information system, has implemented a strict ban on AI-assisted code contributions. Effective October 7, 2026, the project now prohibits any code, documentation, or other contributions generated with the help of large language models (LLMs) or similar AI tools. This policy shift marks a significant change from its previous “human in the loop” approach and reflects growing concerns among open-source maintainers about the impact of AI on project sustainability.
The Policy Change: QEP 428
The ban was formalized through QGIS Enhancement Proposal (QEP) 428, titled “No ai contributions permitted”. Proposed by core developer Nyall Dawson on September 24, 2026, the proposal underwent a community vote on October 5 and was approved and merged into the main repository just two days later, on October 7. The merged policy states plainly: “No LLM, AI or agent assisted code contributions are allowed”, accompanied by provisions for code reversion and sanctions against violators.
This change replaces QEP 408, the project’s previous AI policy which had been in place since January 2026. QEP 408 modeled a “human in the loop” approach inspired by LLVM’s policy, requiring that a human reviewer understand and take responsibility for AI-assisted contributions. Dawson described the old policy as “quite permissive” and acknowledged it “clearly didn’t work” in practice.
Why QGIS Banned AI Contributions
The decision did not come from ideological opposition to AI but from practical project management challenges. According to discussions on the QGIS developer mailing list and the QEP 428 discussion thread, several factors drove the change:
-
Maintainer Burnout: Contributors reported that reviewing AI-assisted pull requests placed an unsustainable burden on maintainers. One reviewer estimated that approximately 50 pull requests had been tagged “AI Assisted” under the old policy, with about 40 of them ultimately closed without being merged. This created a significant backlog of unreviewed or problematic code.
-
Erosion of Trust: The “human in the loop” policy relied on an implicit trust contract between contributors and reviewers. Reviewers could calibrate their expectations based on knowing a contributor’s typical skill level. AI-generated contributions disrupted this calibration, as reviewers could no longer rely on their understanding of the contributor’s abilities to assess code quality.
-
Volume and Quality Issues: AI-generated contributions often arrived as high-volume, seemingly polished code that required significant effort to verify. Maintainers reported spending excessive time checking for subtle errors, security vulnerabilities, or maintenance issues in code that appeared correct at first glance but contained flaws only detectable through deep familiarity with the project.
-
Precedent from Other Projects: QGIS looked to other open-source projects that had already implemented stricter AI policies, particularly GDAL (Geospatial Data Abstraction Library), which bans vibe-coded contributions and limits AI use to specific mechanical tasks like autocompletion.
What the Ban Means for Contributors
Under QEP 428, contributors must now ensure that all submissions are created without any AI assistance. This includes:
– Code written entirely by the human contributor
– Documentation and comments authored without AI generation
– Commit messages and issue descriptions created manually
– Any use of AI tools for brainstorming, research, or learning must not result in direct incorporation of AI-generated content into the submission
The policy does not prohibit contributors from using AI as a learning aid or for general skill development. However, any lines of code, text, or other content that originated from an AI tool must be entirely rewritten by the human contributor before submission. Maintainers reserve the right to revert code and impose sanctions on contributors found violating the policy.
Broader Implications for Open Source
QGIS’s decision reflects a growing tension in the open-source community regarding AI-assisted contributions. While some projects like the Linux kernel and Fedora allow AI use with disclosure requirements, others have moved toward stricter bans. The QGIS ban aligns with projects such as Gentoo (which forbids AI-generated content entirely) and NetBSD (which requires core approval for any LLM-generated code).
For Malaysian developers and technology enthusiasts, this change highlights important considerations:
– Local Impact: Malaysian contributors to QGIS must now adhere to the same strict standards as global contributors, ensuring their submissions are entirely human-generated.
– Career Implications: Developers seeking to contribute to major open-source projects may need to demonstrate their ability to produce work without AI assistance, particularly for high-profile projects like QGIS.
– Policy Awareness: The QGIS ban serves as a case study in how open-source projects are adapting to AI, offering lessons for other Malaysian tech communities and companies navigating similar challenges.
Frequently Asked Questions
Q: Does this ban apply to all types of contributions, including documentation and translations?
A: Yes, the ban covers all types of contributions to the QGIS repository, including code, documentation, translations, commit messages, and issue descriptions. Any submission must be entirely human-generated.
Q: Can I still use AI tools to help me understand QGIS or learn new skills?
A: Absolutely. The policy prohibits incorporating AI-generated content into your submissions, but using AI for personal learning, research, or skill development is encouraged and not restricted.
Q: How will maintainers detect if a contribution used AI assistance?
A: The policy does not rely on detection mechanisms. Instead, it shifts the burden of proof: maintainers can reject contributions that appear suspiciously polished or inconsistent with a contributor’s known skill level, citing QEP 428. Repeat violations may result in sanctions.
Q: Is this ban permanent, or could it be relaxed in the future?
A: The policy is not intended to be “written in stone, forever”. Dawson indicated that exceptions might be considered for specific use cases like IDE autocompletion or translations, but any changes would require community review and voting.
Q: How does this compare to other open-source projects’ AI policies?
A: QGIS now joins a growing list of projects with strict AI bans, including Gentoo and NetBSD. Other projects like the Linux kernel and Fedora allow AI use with disclosure requirements, while Godot permits limited AI for mechanical tasks with disclosure.
Conclusion
QGIS’s ban on AI-assisted contributions represents a pragmatic response to the challenges of maintaining a large, complex open-source project in the era of accessible AI tools. By prioritizing maintainer sustainability and code quality over convenience, the project aims to preserve the collaborative trust that has made it successful for over two decades. While the debate over AI in open source continues, QGIS’s decision underscores that some projects believe the current costs of unregulated AI assistance outweigh the benefits, at least for now. For contributors worldwide, including those in Malaysia, the message is clear: when contributing to QGIS, your work must be entirely your own.