OpenAI to implement invisible watermarks for ChatGPT text in the EU

OpenAI has announced the implementation of an invisible watermark, named textGrain, for text generated by ChatGPT and Codex within the European Union. This measure is intended to comply with the requirements of the EU AI Act, which mandates that machine-generated content be identifiable.

The technology works by adding a statistical signal to the model’s word choices. Rather than inserting hidden characters or symbols, textGrain subtly influences the selection of certain words to create a pattern that a detector can recognize. This process is designed to occur without noticeably altering the quality or meaning of the text. The rollout will apply to all subscription tiers, including free users, but is currently limited to the European Union.

While the system is designed to be resilient to simple copy-pasting, its effectiveness can be diminished through manual editing. Testing indicates that replacing a portion of the words with synonyms can significantly reduce the detection rate. OpenAI plans to release the technology as open-source code at an unspecified future date.