OpenAI will start watermarking ChatGPT’s text in the EU

OpenAI announced Monday in a blog post that it will add an invisible watermark to text generated by ChatGPT and Codex in the European Union. The update complies with the EU AI Act transparency rules that took effect on August 2. The feature will roll out over the coming weeks to eligible users on all plans in the EU.

Developers using OpenAI’s API globally can enable it for select models starting Monday, though it remains off by default. The method, called textGrain, was detailed in a technical report co-written with researchers from the University of Pennsylvania and Yale. It works by subtly shaping word choices to leave an invisible pattern for detectors.

OpenAI noted several limitations during testing. Replacing 10 percent of words with synonyms dropped detection accuracy from about 92 percent to 66 percent. Short passages, math answers, and translated text are also harder to detect. Because of these limits, OpenAI will give initial detector access only to approved researchers and expert organizations.

The company stated that a missing watermark does not prove human authorship.