OpenAI will roll out invisible text watermarks in the EU

OpenAI says it will add invisible text watermarks to eligible ChatGPT and Codex output in the European Union over the coming weeks. The company calls the signal textGrain, a statistical pattern in word choice intended to support provenance checks without changing how text looks.

API customers worldwide can opt in for selected models now, with the option off by default. Detector access will be limited to approved researchers and expert organisations. OpenAI says the EU rollout covers eligible ChatGPT and Codex text, not every response or product surface.

At a one percent false-positive target, OpenAI reports detection of about 80 percent of 200-token psychology samples and 95 percent of 400-token samples. It also reports weaker detection after paraphrasing: 92 to 66 percent with ten percent synonym replacement, and 17 percent after 25 percent replacement.