Proposal Seeks LLM Policy for Drupal Core Contributions

A proposal for a formal large language model policy in Drupal core contributions focuses on managing the impact of AI-assisted work on review workflows. It argues for a harm-reduction and education approach rather than restricting the use of AI tools.

In a post published by Tres Bien Tech, Theodore Biadala argues that the key issue is how AI use affects Drupal’s core contribution process. The proposal highlights increasing use of LLMs alongside inconsistent disclosure, despite an existing requirement.

The proposal outlines a set of directives, including automatic disclosure of AI-generated content, a 180-word limit on generated text, and expectations that code must pass linters and include relevant tests. It also calls for concise commit messages and rejects attribution practices that treat AI tools as co-authors.

The focus is on review efficiency. Biadala argues that review is the main bottleneck in Drupal core development, and that excessive or low-quality AI-generated output can slow the process. The proposed rules aim to reduce noise, improve clarity, and make contributions easier to assess.

The post also distinguishes between suitable and unsuitable use cases. Tasks involving human communication, such as coordination or attribution, require direct human input, while repetitive or structured work may benefit from AI assistance if properly reviewed. The proposal positions policy as a means to guide rather than prevent use.