ChatGPT to add invisible watermarks to text in the EU AMSTERDAM

AMSTERDAM — OpenAI will begin adding an invisible watermark to certain text generated by ChatGPT and Codex in the European Union as part of its compliance with European AI regulations.

The technology, called textGrain, places an invisible statistical signal in the model's word choices. A specialized detector can later assess whether that signal is present. OpenAI says the EU rollout will take place over the coming weeks.

However, the system is not foolproof. Shorter texts are harder to detect, while editing can significantly weaken the signal. In OpenAI testing, replacing 10% of words with synonyms reduced detection from about 92% to 66%.

Initially, access to the detector will be limited to approved researchers and expert organizations. OpenAI also stresses that a watermark does not identify the user or establish how much human editing or creativity was involved.