OpenAI introduces invisible watermarks to ChatGPT texts

In the coming weeks, OpenAI will introduce the textGrain technology to ChatGPT and Codex users in the European Union. This solution is aimed at marking content created by artificial intelligence.

The watermarking mechanism is based on a statistical pattern of word and token selection. The technology is integrated directly into the text structure, without using hidden characters or special spaces.

A special detector will be used to verify authenticity, capable of identifying whether a text was generated or processed by OpenAI models. However, the system's effectiveness may decrease with paraphrasing, translation, or word substitution.

The launch of the technology is directly linked to compliance with EU AI Act requirements for mandatory labeling of content created using artificial intelligence.