OpenAI says it will add an invisible watermark to eligible ChatGPT and Codex text output in the European Union “over the coming weeks.” The mark cannot be seen when you read or copy the text. Instead, a technology OpenAI calls textGrain slightly nudges the model’s word choices to leave a statistical pattern that a detector can pick up later. According to BleepingComputer, API developers worldwide can opt in to watermarking for supported models starting now, but it stays off by default outside the EU, and access to OpenAI’s detector is limited to approved researchers and expert organizations for now.
OpenAI’s own numbers show the limits. In tests on 400-token passages, replacing 10% of words with synonyms cut detection from about 92% to 66%, and replacing 25% of words cut it to 17%. Shorter passages and subjects such as mathematics, where the model has less freedom in word choice, were harder to detect. OpenAI warns that a missing watermark does not prove a human wrote the text, and that a detected watermark does not reveal who generated it, which account was used, or how much of the final work was edited by a person. The company says the watermark does not meaningfully affect output quality in its benchmarks.
The move follows Anthropic’s earlier announcement that future Claude models would watermark text for the same EU market.