OpenAI Will Watermark ChatGPT Text in the EU

OpenAI will start quietly marking the words ChatGPT writes — but only for users in Europe.

The company said Monday it will embed an invisible watermark, called textGrain, into text generated by ChatGPT and Codex in the European Union. The change reaches eligible users on all plans over the coming weeks.

The move is a direct answer to the EU AI Act. The law’s transparency rules, which took effect on August 2, require AI companies to mark generated content in a way other systems can recognize.

The watermark is not a hidden symbol. textGrain uses a secret key to sort the model’s next-word predictions, nudging word choices into a statistical pattern a reader cannot see but a detector can. Because the pattern lives in the words themselves, it survives copy and paste. OpenAI published a technical report on the method alongside researchers from the University of Pennsylvania and Yale.

OpenAI said the mark does not identify the user, and that it found no meaningful change in its models’ performance with the watermark switched on.

The system has limits. In company tests, swapping 10 percent of the words for synonyms dropped detection from about 92 percent to 66 percent. Short passages, math answers and translated text are harder to check. OpenAI cautioned that a missing watermark “does not prove human authorship.” The company is restricting the detector, at first, to approved researchers and expert organizations, so they can help judge how reliable it is.

Outside the EU, nothing changes by default. Developers using OpenAI’s API anywhere in the world can switch the watermark on for select models starting this week — the setting stays off. OpenAI said it is not making text watermarking a global default at launch.

Anthropic got there first. The rival rolled out comparable watermarking for Claude two months earlier, worldwide and mandatory, a move that drew pushback from some users. OpenAI also said it plans to open-source the textGrain implementation.