OpenAI uses textGrain to watermark some EU ChatGPT Codex outputs
Technology
OpenAI is now slipping invisible watermarks into some ChatGPT and Codex-generated text for users in the European Union.
This system, called textGrain, helps the company follow new European Union AI rules by tweaking word patterns so that the watermark is machine-readable and not visible to readers.
OpenAI detector flags watermark not authorship
Right now, only researchers and some organizations get access to the watermark detection tool, while API users worldwide can opt in to certain models.
The detector can indicate whether it finds an OpenAI watermark but doesn't track who wrote what or check for accuracy.
Still, it's a step toward making AI content more transparent, even if you won't notice anything different when you use ChatGPT.