OpenAI adds invisible text watermark in EU

OpenAI is rolling out an invisible text watermark for ChatGPT and Codex outputs in the European Union to comply with the EU AI Act, which requires AI-generated content to be machine-identifiable. The watermark, called textGrain, works by subtly shaping word choices rather than adding a visible symbol, allowing it to persist when text is copied. Detection drops significantly if around 10% of words are replaced, and short passages are harder to flag. OpenAI is limiting detector access to approved researchers initially, and notes a missing watermark does not confirm human authorship. API developers worldwide can opt in today.