LLM tool policy
Rationale
LLMs are changing software development in many ways. What was once scarce, the time and expertise to write software, is now plentiful. This scarcity was a choke point that limited consumption of related resources in open source projects. With that scarcity effectively gone in 2026, the dynamics and economics of a project like GDAL are being disrupted.
A project like GDAL is not simply code. It is also design, review, coordinated refactoring, deprecation scheduling, coordinated communication, distribution, and historical outlook. The plentiful convenience of code writing LLMs do not effectively replace the human maintainers, architects, and documenters doing the jobs that provide much of the value GDAL users actually derive. If they did, LLM contributors would not need to bother to upstream activity to the project at all.
With the explosion of LLM usage in software development, the constrained resource is now "maintenance". It is the time to review your contribution, the time to make it concise, the time to refactor it in to a larger system, and the time to protect the larger software system from disruption, breakage, and performance degradation. Indiscriminate usage of LLMs in open source projects consume maintenance, and the GDAL LLM tool policy attempts to conserve that resource.
Additionally, legal systems across the world (including US and EU) have not definitely determined whether LLM outputs are derived works of training data or if LLM-written code can even be copyrighted by a human. This is despite it being latently extracted and originated from open source software in the first place.
Policy
Contributors can make limited use of LLMs for contributions in GDAL, subject to details mentioned below:
Human contributors must be the primary author(s) of GDAL contributions
All contributions including code, ticket comments, and commit messages should be fully understood by the author(s) submitting them to the project.
Submission of vibe-coded contributions is banned.
LLMs may only be used as an improved auto-completion mechanism, or for repeated tasks (mechanical refactoring) that could potentially be completed with a deterministic algorithm.
Human-coordinated or uncoordinated (OpenClaw, etc) use of agents for submission of contributions to the GDAL repository is banned.
Any LLM usage must be indicated by ticket label, comment, or commit message indication and account for what was written by whom/what.
The contributing human author is ultimately responsible for every line of code, comment, or mailing list interaction they initiate, and all of it is subject to the project's Code of Conduct.
The typical high verbosity of LLM code and text is actively discouraged. More code is more code to maintain. High verbosity contribution (tickets, code, messages, etc) will be seen as indication of LLM-generated content when not labeled otherwise and may be ignored, closed, left unmerged, or removed at maintainers' discretion.
Violations
If a maintainer judges that a contribution does not comply with this policy, they should paste the following response to request changes:
This PR does not appear to comply with our policy on tool-generated content,
and requires additional justification for why it is valuable enough to the
project for us to review it. Please see our developer policy on
AI-generated contributions:
https://gdal.org/community/ai_tool_policy.html
If a contributor fails to rectify their contribution to comply with the policy, maintainers may lock the conversation and/or close the pull request/issue/RFC. In case of repeated violations of our policy, the GDAL project reserves itself the right to temporarily or permanently ban the infringing person/account.
Mitigation
The GDAL Sponsorship Program is one way your organization can help buffer the cost and disruption of LLMs in keystone projects such as GDAL. The constrained resource is maintenance, not adding more code/capability. The Sponsorship Program financially supports operation of GDAL as an ongoing open source software project, and without it, much of the activity GDAL users take for granted would simply not happen.