Prime Highlights-
- OpenAI adds invisible textGrain watermark to ChatGPT and Codex text in the European Union, meeting EU AI Act transparency rules.
- Watermark reaches eligible ChatGPT and Codex users on all EU plans, while developers worldwide can opt in via API.
Key Facts-
- OpenAI watermark gently steers word choices, travels with copied text and keeps user identity private.
- Approved researchers and expert organizations get first detector access to test reliability and responsible use.
Background-
ChatGPT and Codex users in the European Union will soon see an invisible watermark on AI-generated text, as OpenAI moves to meet the transparency rules of the EU AI Act, the company announced in a blog post.
The rollout reaches eligible users on every ChatGPT and Codex plan across the EU over the coming weeks. Developers worldwide can opt in through the API for select models. OpenAI keeps the EU as the focus of this launch.
The watermark works by gently steering the model’s word choices, creating a pattern that a detector can read. Since the pattern sits in the words themselves, it stays with the text when users copy and paste it. OpenAI said the watermark keeps user identity private and that model performance held steady during testing.
OpenAI also published a technical report on its textGrain method, co-written with researchers from the University of Pennsylvania and Yale. A secret key sorts the model’s next-word options, and a detector with the text and the key can confirm AI-generated content.
In the first phase, approved researchers and expert organizations will receive detector access and help OpenAI test reliability and guide responsible use. The company said a watermark shows an OpenAI system generated or processed part of a passage, and human judgment, editing and creativity add to that work. Anthropic already applies a similar watermark to Claude text across the world.