[DRAFT] Policy on the use of AI tools in W3C Standardization work
Status: This is a draft document for discussion and review; it is not (yet) a W3C policy. Feedback welcome through GitHub in the TPAC 2026 breakout session proposal.
Purpose
This policy sets expectations and requirements for any use of Artificial Intelligence (AI) tools for standards work at W3C. It builds on the Use of Large Language Models in Standards Work document published by W3C’s Advisory Board.
In general, a variety of tools are used in W3C to help with the development of technical specifications and best practices, including for authoring, versioning, validation, analysis, cross-linking, and automation. These tools are valued when they improve collaboration. Due to their ability to generate content at will, AI tools may hinder collaboration when their usage is unsanctioned or without guardrails.
This policy is based upon the principle that every W3C contribution must be attributable to a human contributor who has reviewed and approved its content before publication, understands its substance, and accepts responsibility for its relevance, meaningfulness, correctness, quality, security, scope, provenance, and compliance with other applicable W3C policies. AI tools may be used to assist with a contribution, but may not be treated as contributors or act as unsupervised participants in W3C work.
Policy
Do not use AI tools unless you are willing to assume full responsibility
Participants are fully responsible for their contributions. This is specifically true for content (partially) outputted by AI tools and for which it may be tempting, but inappropriate, to delegate responsibility. Among other things, participants are required to check that the content complies with other W3C policies, including licensing requirements and the Code of Conduct.
Keep a clear purpose to your proposals
AI tools can generate a lot of content. Contributors should be mindful that reviewers are humans, and be respectful of their time and effort. People should not use AI tools to generate lengthy, redundant, or high-volume proposals, comments, or issue activity without first establishing that the material is necessary and welcome. They should not use AI tools to overwhelm discussions with repeated proposals. Clarity and brevity facilitate collaboration in a global community.
Establishing that the material is necessary and welcome includes reviewing contribution guidelines for the project, which may set additional requirements. For example, a group may require that substantive updates to a specification link back to an issue where these changes were discussed and approved, or set more constrained requirements regarding usage of AI tools to generate specification updates.
Disclose AI usage
When the substance of a contribution was generated by an AI tool, label the content as such. This helps reviewers evaluate the content.
Embrace thoughtful discussions
A human reply calls for a human reply. Contributors must be the ones to reply to feedback they receive.
Similarly, participants must not let an AI tool autonomously reply to, or otherwise engage in a discussion with other participants.
Do not impose AI tools to other participants
Groups may adopt tools to improve collaboration. Individual participants must not impose tools to others without first establishing that the tool is welcome. This applies to AI tools that may fully or partially record a meeting, for example for transcription or summarization. Also see Automated Transcripts in the Guidebook.
Consider impacts
Contributors are encouraged to be conservative about energy usage and, if they do choose to use AI, to favor specialized or smaller AI tools and models over large and general-purpose models to reduce the environmental impact of their contributions. See the Responsible Consumption and Production goal of the United Nations Sustainable Development Goals.
Handling policy violations
People with specific roles at W3C, including specification editors, group chairs, test leads, project maintainers and reviewers, are encouraged to act on content that fails, or appears to fail, any of the above requirements. Possible actions include dismissing the content, blocking an underlying account on GitHub (especially when this account is not associated with a known W3C participant), or following the procedures for handling incidents relating to Code of Conduct violations for mailing lists or Groups if needed.