OpenAI will start watermarking ChatGPT’s text in the EU
TechCrunch Aditya Mehta ● Covered by 4 sources
OpenAI is adding invisible watermarks to ChatGPT and Codex text in the EU. It’s for the EU AI Act, and the mark travels with copied text.
Based on reporting by TechCrunch, Aditya Mehta — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI said Monday it will start embedding an invisible watermark in text made by ChatGPT and Codex for users in the European Union. The move is meant to satisfy the EU AI Act, whose transparency rules kicked in on August 2 and require AI-generated content to be marked so other systems can identify it.
The rollout will happen over the next few weeks and apply to eligible ChatGPT and Codex users on all plans, but only in the EU. Developers using OpenAI’s API can switch it on for select models anywhere in the world starting today. By default, it stays off, and OpenAI is not making text watermarking a global default at launch.
This is not a visible stamp. OpenAI’s method works by nudging word choices in a way people cannot see but a detector can. Because the pattern is built into the words themselves, it can survive copy and paste. OpenAI says the watermark does not identify the user, and that it saw no meaningful change in model performance with the system switched on.
The company also published a technical report on the method, called textGrain, co-written with researchers from the University of Pennsylvania and Yale. The report describes using a secret key to sort next-word predictions, then stacking hundreds of those tiny shifts so a detector can flag AI-generated text from the text alone.
OpenAI is also being blunt about the limits. Its tests found that replacing 10% of words with synonyms cut detection from about 92% to 66%. Short passages, math answers and translated text are harder to catch. So OpenAI is giving detector access first to approved researchers and expert organizations, and warning that a missing watermark does not prove human authorship.
My take — AI-written commentary, not fact-checked reporting
This is the right kind of boring. If AI companies are going to pump out text at scale, they should also help label it at scale, even if the result is a clever bit of word-shuffling rather than a neon sign. The real test is whether rivals follow suit or quietly hope everyone else does the honest thing first.
Read more about this at: TechCrunch
