OpenAI is adding invisible watermarks to ChatGPT text in the European Union. The company announced the move on October 5. The system is called textGrain. It embeds a hidden statistical signal in the model’s word choices. Readers cannot see it. Special detector software can spot the pattern. The watermark will roll out to eligible ChatGPT and Codex users in the EU over the coming weeks. It applies to all plans in the bloc. The move implements the EU AI Act’s transparency rules. The Act requires AI-generated content to be machine-readable. Developers worldwide can switch on watermarking for select models through the API. The setting stays off by default outside the EU. OpenAI opened applications for a watermark detector. Only approved researchers and expert organisations can use it. The detector will not be public at launch. OpenAI says detection works best on longer passages. At a 1 percent false positive rate, it spots watermarks in about 80 percent of 200-token passages. For 400-token passages, the rate is about 95 percent. The company warns editing, translation and short passages weaken detection.