ChatGPT EU Watermark: How OpenAI's textGrain Works

Your ChatGPT answers in Europe are about to carry a hidden signature. On October 5, 2026, OpenAI said it will add an invisible watermark to text generated by ChatGPT and Codex in the European Union, TechCrunch reports, to comply with the EU AI Act transparency rules in force since August 2.

Who gets it, and when

• EU only for ChatGPT and Codex, rolling out over the coming weeks to eligible users on all plans
• API: developers anywhere can switch it on for select models from October 5—off by default
• Not a global default at launch

How textGrain works

The watermark isn’t a visible symbol. Using a secret key, it subtly nudges the model’s next-word choices; hundreds of such nudges form a pattern a detector can spot from the text alone. OpenAI published a technical report on the method, called textGrain, co-written with researchers from the University of Pennsylvania and Yale. Because it lives in the words, it survives copy-paste; OpenAI says it doesn’t identify the user and saw no meaningful change in model performance.

Limits OpenAI admits

• Swapping 10% of words for synonyms cut detection from about 92% to 66%
• Short passages, math answers and translated text are harder to detect
• Detector access goes first to approved researchers and expert organisations only
• A missing watermark “does not prove human authorship”

Anthropic announced a worldwide watermark for Claude text two months earlier, which drew pushback from some users.

Rocks take

For EU students, marketers and newsrooms, assume ChatGPT copy is now traceable by OpenAI’s detector partners—but not by your teacher’s random “AI checker.” Translation weakening the signal matters a lot in a multilingual market.

FAQ

Can I turn it off in the EU? OpenAI hasn’t described an opt-out for EU ChatGPT users.

Does it reveal who wrote the prompt? No, according to OpenAI.