OpenAI will add an invisible watermark to text generated by ChatGPT and Codex in the EU to comply with the EU AI Act's transparency rules, rolling out over the coming weeks to eligible users on all plans.

The watermark, which is off by default for API developers worldwide, works by subtly shaping word choices and does not identify the user.

OpenAI noted that editing can remove the watermark and that a missing watermark does not prove human authorship.