OpenAI announced it will add invisible watermarks to ChatGPT text in the EU to comply with regulations, but will let API users globally opt out.

The move is part of the EU AI Act, which requires AI providers to label generated text in a machine-readable format. OpenAI's solution, called textGrain, embeds an invisible statistical signal in the model's word choices, similar to Claude's SynthID watermark.

OpenAI says textGrain matched or beat other approaches in internal tests, including Google's SynthID for text. The company also plans to release the technology as open source so others can build on it. Detection rates depend heavily on text length and subject matter, with the detector set to a target false-positive rate of 1 percent.

At 400 tokens, the detector identifies about 94 percent of watermarks in psychology passages but only about 60 percent in math passages.

OpenAI says editing the text also makes the watermark harder to detect, with replacing 25 percent of the words pushing detection below 20 percent, even in 400-token passages.

That weakness could work in OpenAI's favor, as users who don't want their ChatGPT use detected might switch to open-weight models.

"TextGrain works much like Claude's SynthID watermark, which uses Google's open-source technology," said OpenAI. The company says watermarking doesn't hurt output quality, citing tests of its frontier model Astra. It reports no significant performance differences with watermarking on or off across eight benchmarks, including GPQA Diamond, BrowseComp, and DeepSWE.

OpenAI also says a detected watermark reveals nothing about how much human creativity or editing went into the text. It doesn't establish ownership, assign responsibility, identify a user, or show whether the text is accurate.

Failing to detect a watermark doesn't prove a human wrote the text, either, as the passage could be too short, have been edited or translated, or come from a model the detector doesn't support.

OpenAI will limit detector access for now, granting it only to selected researchers and specialist organizations. They can apply through an application form. The company plans to expand access "when we believe results can be interpreted responsibly." Existing verification tools for images and audio remain publicly available, including openai.com/verify and the Content Provenance API.

Source: thedecoder