OpenAI will add invisible watermarks to ChatGPT and Codex text in the EU to comply with the AI Act

OpenAI said Monday that it will add an invisible watermark to text generated by ChatGPT and Codex in the European Union. The move is meant to comply with the EU AI Act's transparency rules, which took effect on August 2 and require AI companies to mark generated content in a way other systems can identify. The watermark will roll out over the coming weeks to eligible users on all plans, but only in the EU. API developers anywhere can turn it on for select models starting now, though it is off by default. OpenAI said it is not making watermarking a global default at launch. The method, called textGrain, uses a secret key to subtly shape the model's word choices. This leaves a statistical pattern that readers cannot see but a detector can pick up, and the pattern survives copying and pasting. OpenAI says the watermark does not identify the user and did not meaningfully change model performance. A technical report was co-written with researchers from the University of Pennsylvania and Yale. OpenAI acknowledges limits. In 400-token passages, replacing 10 percent of words with synonyms cut detection from about 92 percent to 66 percent, and replacing 25 percent cut it to 17 percent. At first, access to the detector is limited to approved researchers and expert organizations. The Verge reports that OpenAI says textGrain matched or exceeded other approaches, including Google DeepMind's SynthID for text.