OpenAI introduces invisible watermarks for text in ChatGPT and Codex
OpenAI has announced the launch of its textGrain invisible text watermarking system in ChatGPT and Codex for users in the European Union.
The change is prompted by the entry into force of the European Union's Artificial Intelligence Act, which requires creators of generative models to ensure machine detection of generated content. In the coming weeks, invisible watermarks will begin rolling out in responses across all subscription plans in the EU. In addition, API customers worldwide have been given the option to voluntarily enable the feature for a number of models, while access to the watermark detector is initially being provided only to accredited researchers and relevant organizations.
The textGrain method does not rely on hidden characters or spaces; instead, it introduces a statistical pattern during the word- and token-selection phase of the neural network. According to OpenAI's internal benchmarks, the procedure barely affects generation quality. At the same time, the developers note the technology's vulnerability: in a 400-token text, replacing 10% of the words reduces the detector's accuracy from 92% to 66%, and replacing 25% of the words drops it to 17%. The detector also performs worse on short passages and in exact disciplines where phrasing choices are limited.
The company emphasized that the textGrain watermark does not identify the user, does not reveal the text of their prompts, does not assess the extent of human involvement in the writing, and does not serve as proof of the factual accuracy of the information.
What it means
For ChatGPT and Codex users in the European Union, generated text across all plans will begin to be invisibly watermarked in the coming weeks, while the company's global API customers can enable watermarking at their own discretion.
Sources
- Kod.ru: OpenAI внедряет невидимые водяные знаки для текста в ChatGPT и Codex media outlet
- iXBT.com: OpenAI запускает в ChatGPT невидимую маркировку текстов media outlet
- 3DNews: ChatGPT и Codex начнут помечать ответы невидимыми «водяными знаками», но не у всех media outlet
- Hi-Tech Mail: В текстах ChatGPT появятся водяные знаки для обнаружения ИИ media outlet
This article was prepared with the help of AI and translated from the Russian original. Spotted an error — let us know.
Published 6 October 2026, 11:31 · No updates · Russian original




Comments on this story live under its post in our Telegram channel. Comment on Telegram