OpenAI introduces invisible text watermarking for ChatGPT and Codex in EU

OpenAI is preparing to add invisible watermarks to text generated by ChatGPT and Codex in the European Union, BleepingComputer reported on 5 October 2026. BleepingComputer, The Verge, and The Next Web all reported that the company is deploying the feature across its tools in the region. [1][2][3][4][5][6]

The watermark will not be visible when reading or copying text, according to BleepingComputer. OpenAI explained that its textGrain technology slightly alters the model's choices of words to create a statistical pattern that can be detected later. The company stated that the watermark will be added to eligible output in the coming weeks. [1]

Application programming interface developers worldwide can opt in to watermarking for supported models starting on 5 October 2026, but the feature remains disabled by default, BleepingComputer reported. OpenAI is also opening applications for its watermark detector tool. [1]

ChatGPT was originally released on 30 November 2022 by OpenAI and uses large language models to generate responses. Statistical text watermarking for large language models modifies model outputs so they can be identified later as generated by artificial intelligence. [7][8]