The open-source project LLVM has introduced a new policy allowing AI-generated code in contributions, provided humans review and understand the submissions. This 'human in the loop' approach ensures accountability while addressing community concerns about transparency. The policy, developed with input from contributors, balances innovation with reliability in software development.
LLVM, a foundational collection of compiler and toolchain components used in projects like Clang, Rust, Swift, and the Linux kernel, has adopted a policy on AI tool use in contributions. Published on January 22, 2026, the guidelines permit developers to employ any AI tools but emphasize full accountability for the submitted work.
Under the policy, contributors must disclose the AI tool used, either in the pull request description, commit message, or authorship details. They are required to review and comprehend their submissions, confidently justifying them during reviews and ensuring they merit a maintainer's attention. The rules clarify that violations will be handled according to existing community processes.
The development process involved extensive community engagement. A LLVM member highlighted discrepancies between the project's AI handling, code of conduct, and actual practices, referencing a notable pull request discussed on Hacker News where AI use was admitted post-submission but not initially disclosed.
LLVM maintainer Reid Kleckner spearheaded the effort. His initial draft, inspired by Fedora's AI policy, proposed restrictions such as limiting newcomers to 150 lines of non-test code. After feedback from community meetings and forums, the final version shifted to more explicit requirements, focusing on review readiness and question-answering ability rather than vague ownership clauses.
The updated AI Tool Use Policy is now available on LLVM's documentation site, including examples of acceptable AI-assisted work and violation guidelines. This move aligns LLVM with other open-source initiatives adapting to AI's growing role in development.