OpenAI will watermark ChatGPT outputs by default—but only in the EU
Ars Technica Samuel Axon ● Covered by 6 sources
OpenAI will watermark ChatGPT text by default in the EU. It’s to satisfy the EU AI Act, but the marks may still be easy to dodge.
Based on reporting by Ars Technica, Samuel Axon — 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 is turning on automatic watermarking for ChatGPT text in the European Union. Outside the EU, the feature will still exist, but people will have to opt in.
The change is meant to line up with the EU AI Act, which took effect in August and requires AI-made content to be marked in a way another tool can detect. That sounds clean on paper. In practice, the source of the problem remains the same: no watermarking system is fully reliable yet.
OpenAI’s version is called textGrain. The company says it is proprietary and has published a technical paper on how it works. The basic idea is familiar. The model nudges word choices in ways a human reader should not notice, but a detector with the right key can pick up.
That puts OpenAI alongside existing approaches such as SynthID and the C2PA project, both of which are already known to be imperfect and relatively easy to get around with basic know-how. OpenAI says access to its detector will go to a limited set of researchers and organizations at first, with a request process for adding others over time.
My take — AI-written commentary, not fact-checked reporting
This is the usual EU story: regulation arrives, and the industry responds with a detection system that is useful mainly until someone bothers to evade it. Watermarking is better than pretending AI text is magic, but it is still compliance theater with a technical appendix. OpenAI gets points for trying; nobody should confuse that with solving the problem.
Read more about this at: Ars Technica
