LLVM Establishes Formal AI Tool Use Policy with Human-in-the-Loop Requirements
Key Takeaways
- ▸Contributors must read and review all LLM-generated code before requesting human review; the contributor remains fully accountable
- ▸Automated agents taking action without human approval are banned, including GitHub's @claude agent and automated review tools that publish without review
- ▸AI tool usage must be transparently labeled in commit messages, PR descriptions, or standard authorship locations to facilitate review and develop community best practices
Summary
The LLVM Project has published a comprehensive policy governing how contributors can use AI tools like large language models and agents in their work. The policy mandates that contributors must review all LLM-generated code and text before submission, maintain full accountability for their contributions, and transparently label submissions containing substantial AI-generated content using commit message trailers such as 'Assisted-by'.
The policy introduces several key restrictions, including banning autonomous agents that take action without human approval (specifically citing GitHub's @claude agent) and prohibiting the use of AI tools to fix issues labeled 'good first issue,' which are reserved as learning opportunities for new contributors. The policy explicitly states that contributions must represent a net positive for the project—worth more time than required to review them.
LLVM frames the restrictions within the concept of 'extractive contributions,' where unreviewed LLM output is sent to maintainers, extracting their time and energy in review and design work. The project aspires to be welcoming to newcomers while preventing AI automation from undermining both code quality standards and the educational value of contributing to open source.
- Using AI tools to fix 'good first issue' labeled issues is explicitly forbidden to preserve learning opportunities for new contributors
- The policy reflects concern about 'extractive contributions' where low-effort, high-review-burden LLM outputs impose unfair costs on volunteer maintainers
Editorial Opinion
LLVM's policy represents a mature, nuanced approach to AI tool governance in open source—neither rejecting AI tools outright nor allowing unfettered automation. By mandating human review and accountability while banning autonomous agents, LLVM acknowledges that AI tools can improve efficiency when used thoughtfully, but protects both code quality and the social contract of open source. As more projects adopt AI-assisted development, this policy may serve as a template for balancing innovation with maintainer sustainability and contributor education.
