OpenAI Adds Invisible Watermark to ChatGPT in Europe

OpenAI will start adding an invisible watermark to text generated by ChatGPT and Codex in the European Union, the company announced Monday. The move is to comply with the EU AI Act, which requires AI companies to mark AI-generated content so other systems can identify it. The watermark isn't a visible symbol; it's a subtle pattern in the model's word choices that a detector can pick up. Because it's embedded in the words, it travels with the text when copied and pasted.

The watermark will roll out over the coming weeks to eligible ChatGPT and Codex users on all plans, but only in the EU. Developers using OpenAI's API worldwide can turn it on for select models starting today, but it's off by default. OpenAI says the watermark doesn't identify the user and doesn't affect model performance. The company also published a technical report called textGrain, co-written with researchers from the University of Pennsylvania and Yale, explaining how the method works.

However, the watermark isn't foolproof. OpenAI's tests show that replacing just 10% of words with synonyms can drop detection from about 92% to 66%. Short passages, math answers, and translated text are also harder to detect. Because of these limits, OpenAI is only giving detector access to approved researchers and expert organizations for now. The company also warns that a missing watermark doesn't prove human authorship—the text could be too short, heavily edited, or from another AI.

The announcement comes two months after Anthropic said it would watermark text from Claude worldwide, a move that drew backlash from some users who felt they were the creative force behind the output. OpenAI had built a text watermark before but held off, partly fearing users would switch to rivals. Now, with the EU rules in effect, OpenAI is joining Anthropic, Google, Meta, and Microsoft in committing to the EU's code of practice on AI-generated content.