OpenAI to watermark ChatGPT and Codex text in the EU under AI Act rules

OpenAI announced it will begin watermarking text generated by ChatGPT and Codex for users in the European Union to meet the EU AI Act’s transparency obligations. The feature will roll out over the coming weeks to eligible users on all plans within the EU.

Developers using OpenAI’s API anywhere in the world can enable the watermark for select models starting immediately, though it will be off by default. OpenAI emphasized that it is not making text watermarking a global default at launch.

The watermark is not a visible symbol; instead, it embeds subtle patterns in the model’s word choices that a detector can identify. Because the watermark is embedded in the text itself, it persists when content is copied or pasted.

OpenAI noted that the watermark does not identify individual users and reported no measurable change in model performance when the feature is active. The company also published a technical report describing the method, called textGrain, which was co-authored with researchers from the University of Pennsylvania and Yale.

The company acknowledged limitations: editing can degrade detection. In one test, replacing 10% of words with synonyms reduced detection rates from about 92% to 66%. Short passages, math answers, and translated text were also flagged as harder to detect reliably.

OpenAI said it is initially providing detector access only to approved researchers and expert organizations to evaluate reliability and responsible use. It also cautioned that the absence of a watermark does not confirm human authorship, as text may be too short, heavily edited, or generated by another provider’s system.

The announcement follows a similar move by Anthropic to watermark text generated by Claude, though that effort drew criticism from some users who argued they provided key instructions and context.