Article
ChatGPT text in the EU is getting an invisible mark. For now, only people OpenAI approves can check it
On 5 October 2026 OpenAI announced that over the coming weeks ChatGPT and Codex text in the European Union will carry an invisible statistical watermark called textGrain, because Article 50 of the EU AI Act requires it. According to OpenAI's report as described by ActuIA, the detector catches about 95 percent of 400-token English passages on psychology (TechCrunch gives about 92 percent for unedited text), but only 66 percent once 10 percent of the words are swapped for synonyms, and between 42.2 percent (Romanian) and 69.0 percent (Spanish) in other EU languages. At first the detector goes only to researchers and expert organisations OpenAI approves. For developers using the API, the mark stays optional worldwide.
Every piece of text ChatGPT writes for a user in the European Union will soon carry a mark that neither the writer nor the reader can see, and for now OpenAI alone decides who gets to check it.
This touches anyone who uses ChatGPT at work, at university or for email. The key point: the mark does not prove who wrote a text, and its absence does not prove a human did. OpenAI says so itself.