Signals/ ChatGPT & OpenAI/ Reel /
ChatGPT Is Hiding a Secret Signature in Its Text
OpenAI is rolling out “textGrain”, an invisible watermark for ChatGPT and Codex text, in the EU (to comply with the EU AI Act).
How it works: ChatGPT nudges its word choices into a pattern you can’t see, but a detector can. Copy and paste doesn’t remove it.
Before you panic:
- EU only for now
- It contains no information about who you are
- There’s no public checker yet (only approved researchers)
And it’s fragile: in OpenAI’s own tests, swapping 25% of the words dropped detection from about 92% to 17%.
Even OpenAI says a match doesn’t prove AI wrote the whole thing. If AI helped you write it, just say so.
