ChatGPT users, your AI-written text will soon carry a hidden watermark

For years, AI tools have made it easy to generate everything from images and audio to human-like text. While AI-generated images, videos, and audio can often be identified using dedicated detection tools, AI-written text has largely escaped the same level of scrutiny. ChatGPT, in particular, has become one of the most widely used tools for generating text that can closely mimic human writing. Now, OpenAI says it plans to change that by introducing watermarks for text generated by ChatGPT and Codex.

The company says its text watermarking program will first roll out in Europe, owing to the EU AI Act, which requires AI providers to make generated text identifiable in a machine-readable way. Over the next few weeks, OpenAI will add an invisible watermark to eligible ChatGPT and Codex text output in the EU, and API customers globally will also be able to opt into text watermarking for select models.

OpenAI is also accepting applications for its text watermarking detector, with access initially limited to approved researchers and expert organizations to help test the technology.

How does text watermarking work on ChatGPT-generated text

OpenAI explains that its watermarking technology, “textGrain,” adds a statistical signal to the model’s word choices. The detector then looks for that signal to verify if the text was generated using ChatGPT or Codex.

While the company says textGrain matched and even exceeded the performance of all the approaches it tested, it doesn’t always work as intended.

“Detectors can make two kinds of errors: they can report a watermark where none is present — a false positive — or miss a watermark that is present — a false negative,” OpenAI noted in its blog. The company also said that shorter text is harder to detect, and that editing can weaken the watermark.

Moreover, the watermark detection system cannot determine how much a human contributed to the text, who owns it, who created it, or whether the information it contains is accurate.

For now, OpenAI will introduce text watermarking to all users across all plans in Europe. The system is not going global just yet. The company says its regional approach will give it room to learn from real-world feedback.