OpenAI will watermark ChatGPT text in the EU

On October 5 OpenAI said it will add an invisible watermark to text produced by ChatGPT and Codex in the European Union. The system, called textGrain, is designed to meet the EU AI Act transparency rules that took effect on August 2, and rolls out over the coming weeks to eligible users on all plans in the EU.

textGrain adds no hidden characters. It gently shapes the model's word choices so that a statistical pattern sits in the words themselves. Readers cannot see it, but a detector holding a secret key can. Because the signal lives in the words, it survives copy and paste, and OpenAI says it does not identify the user.

Rewriting erodes the signal. Swapping 10% of words for synonyms drops detection from about 92% to 66%, and heavier paraphrasing should push it lower. The watermark is a deterrent and a way to flag large-scale content farms, not proof of authorship.

Images offer thousands of pixels to hide a mark; in text every word carries meaning, and pushing word choice too hard hurts quality. textGrain balances this by nudging choices between near-synonyms.

From October 5, API developers anywhere can switch the watermark on, but it stays off by default outside the EU ChatGPT rollout.

The AI Act sets transparency duties for general-purpose models and the companies deploying them, starting with machine-readable marking of AI-generated or altered content. Images and audio can carry marks in metadata or pixels; text can only carry them in the word choices themselves, which is the approach textGrain takes.

The AI Act requires machine-readable marking of AI-generated content. Text has been the hardest medium to mark, and OpenAI's move sets a bar for other providers.

Publishing raw AI text is becoming more traceable, at least in the EU. The durable answer is the same as for search visibility: make content original through expertise, your own data and real editing.