Guidelines for AI-generated spec contributions

[zcorpan]

Should there be a guideline for accepting spec contributions that appear to be, or are disclosed as being, AI-generated?

Example: whatwg/dom#1446

[michaelchampion]

Yes there should be a guideline. As in the DOM example, it's fine (maybe even good practice for less experienced people?) for a HUMAN agent to use AI tools to make sure the logic, English grammar, code examples, etc. are clear. But the human should disclose up front that they got help from an AI, and maybe even the actual prompt used if requested.

I don't think you can accept contributions from an AI agent even if a human has reviewed it. For one thing, the IPR provenance would be muddled at best.

[zcorpan]

I think it will become increasingly difficult to tell the difference between a human using AI tools and an AI agent. Also, what is the difference for IPR?

[michaelchampion]

If WHATWG does pursue guidelines for AI spec contributions, this https://redmonk.com/kholterhoff/2026/02/26/generative-ai-policy-landscape-in-open-source/ might be a useful resource

I compiled and analyzed the generative AI policies of 32 open source organizations, including foundations like the Linux Foundation, Apache, and Eclipse, as well as individual projects like the Linux Kernel, Gentoo, curl, and Matplotlib. The resulting visualization maps the current policy landscape across several dimensions: overall stance (permissive, ban, or undecided), primary concerns (quality, copyright, and ethics), disclosure requirements, and adoption timeline.