OpenAI to watermark ChatGPT and Codex text in the EU

OpenAI will begin adding invisible watermarks to eligible text generated by ChatGPT and Codex in the European Union over the coming weeks, the company announced.

The change is part of OpenAI’s response to the EU AI Act, which requires providers of generative AI systems to make AI-generated text identifiable in a machine-readable format. Watermarking will roll out to eligible ChatGPT and Codex users across all plans in the EU.

API customers worldwide can opt in to text watermarking starting today for select models. It will remain disabled by default. OpenAI is also accepting applications for access to its watermark detector, although initially only approved researchers and expert organizations will be able to use it. The company eventually plans to make the watermarking technology open source.

OpenAI’s system, called textGrain, adds an invisible statistical signal by subtly influencing the words a model chooses. According to the company’s tests with GPT-6 Astra, watermarking produced no meaningful overall difference in model performance across its benchmarks.

The technology has significant limitations. Short passages are harder to identify, and detection is substantially less reliable for subjects such as mathematics, where models have less freedom in how they phrase an answer. At a 1% false-positive rate, OpenAI detected watermarks in about 80% of 200-token passages and about 95% of 400-token passages for less constrained content.

Editing the text can also quickly weaken the watermark. In OpenAI’s tests of 400-token passages, replacing 10% of the words with synonyms reduced detection from about 92% to 66%. Replacing 25% of the words brought it down to 17%.

OpenAI says the watermark does not identify the user or reveal their account, prompts or conversations. It also cannot determine whether a text is accurate, and failing to detect a watermark does not prove that a passage was written by a human.