OpenAI Adds Watermarks to ChatGPT and Codex Text for EU Users

OpenAI announced that within the next few weeks it will add invisible watermarks to text generated by ChatGPT and Codex for EU users, to meet the EU AI Act’s requirement for machine-readable marking. Earlier reports said API users could turn the feature on themselves worldwide, off by default; later reports said it is currently only for EU users and will take effect automatically in the coming weeks. The latest reports say the watermark is called textGrain, a statistical pattern in word choice rather than a visible label, and will roll out to all plans within the next few weeks, though for now it is limited to the EU. OpenAI says the watermark does not identify users, accounts, or prompts, and the detector is not public—only approved researchers can apply to use it. The feature has not yet gone live, and OpenAI has not announced a specific effective date.

Generated by AI from reports · Updated 4 hours ago