OpenAI will hide a signature in every word ChatGPT writes in the EU.
OpenAI said on Monday it will add an invisible watermark called textGrain to ChatGPT and Codex text. Swapping one word in ten dropped the detector's accuracy from 92 percent to 66 percent. The mark reaches EU users over the coming weeks, and only approved researchers get the detector.
Why it mattersIf you paste ChatGPT text into an essay or report, a hidden mark may travel with it. A missing mark proves nothing, so this will not settle who really wrote something.
Read in the app — free, in 9 languages
Related stories
Utah let an AI write a first prescription with no doctor signing off.2026-10-06Norway moves to ban AI glasses from playgrounds and changing rooms.
2026-10-06Under oath, four AI firms would not say if AI could be catastrophic.
2026-10-06Anthropic handed a user's diary entry to the police.
2026-10-05Two senators want prison time for bosses whose AI agents hack.
2026-10-05