OpenAI will watermark ChatGPT and Codex text in the EU

OpenAI says the EU AI Act "requires generative AI providers to make generated text identifiable in a machine-readable way".

The EU AI Act requires generative AI providers to make generated text identifiable in a machine-readable way.

OpenAI says: "Starting today, API customers globally will be able to opt in to text watermarking for select models. Text watermarking will remain off by default in the API."

Starting today, API customers globally will be able to opt in to text watermarking for select models. Text watermarking will remain off by default in the API.

OpenAI says: "Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union."

Over the coming weeks, we will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union.

OpenAI says it will introduce text watermarking to eligible ChatGPT and Codex users across all plans in the EU only, and is "not making text watermarking a global default at launch".

Over the coming weeks, we will introduce text watermarking to eligible ChatGPT and Codex users across all plans in the EU only. We are not making text watermarking a global default at launch.

OpenAI is opening applications for its text watermark detector, with access initially "limited to approved researchers and expert organizations".

We’re opening applications to access our text watermark detector. Access will initially be limited to approved researchers and expert organizations that can help us evaluate and improve the technology.

OpenAI names its text watermarking technology textGrain, which "adds an invisible statistical signal to the model’s word choices".

Our text watermarking technology, textGrain, adds an invisible statistical signal to the model’s word choices.

At a target false positive rate of 1%, OpenAI says its detector identified watermarks in about 80% of 200-token passages, compared with about 95% of 400-token passages, for content such as psychology, with substantially lower rates for content such as mathematics.

At a target false positive rate of 1%, our detector identified watermarks in about 80% of 200-token passages, compared with about 95% of 400-token passages, for content such as psychology. Detection rates were substantially lower for content such as mathematics, where there is less flexibility in word choice.

In an evaluation of 400-token passages, OpenAI says replacing 10% of words with synonyms reduced detection from about 92% to 66%, and replacing 25% of words reduced it to 17%.

In an evaluation of 400-token passages, replacing 10% of words with synonyms reduced detection from about 92% to 66%. Replacing 25% of words reduced it to 17%.

OpenAI says that in its evaluations textGrain "matched or exceeded the performance of other approaches we tested, including SynthID for text".

In our evaluations, textGrain matched or exceeded the performance of other approaches we tested, including SynthID for text.

OpenAI says it plans to make the technology available in open source.

We also plan to make the technology available in open source so that others can build on it.

The textGrain technical report lists authors from the University of Pennsylvania (Xiang Li, Qi Long), Yale University (Garrett Wen, Xiaohong Chen) and OpenAI, and is dated 5 October 2026.

Xiang Li University of Pennsylvania Garrett Wen Yale University Xiaohong Chen Yale University Qi Long University of Pennsylvania Arzav Jain OpenAI ... October 5, 2026

The report says the method couples token generation to keyed randomness through an optimal transport problem with Gumbel-based costs and KL regularization, and that the detector requires only the generated text and the secret key.

The method couples token generation to keyed randomness through an optimal transport problem with costs based on Gumbel random variables and Kullback–Leibler (KL) regularization. ... The detector requires only the generated text and the secret key and does not need to know the budget used during generation.