OpenAI Adds Invisible Watermarks to ChatGPT Text in Europe

OpenAI has started adding invisible watermarks to text generated by its AI tools, ChatGPT and Codex, but only for users in the European Union. The watermark, named textGrain, is embedded in the text and can be detected by machines, not by people. It's designed to meet the EU's AI Act, which requires transparency about AI-generated content. OpenAI says it's not making this a global default yet, so users outside the EU won't see it for now.

So what does this mean for you? If you're in the EU, the text you get from ChatGPT will contain a hidden marker. This could help identify AI-written content, which is useful for teachers, employers, or anyone worried about misinformation. But OpenAI admits the watermark isn't foolproof—it can't guarantee detection, and it doesn't prove who wrote the text or how much a human was involved. Also, API customers worldwide can choose to add watermarks to their AI outputs, giving businesses more control.

OpenAI is also giving approved researchers access to a detector tool to check for watermarks, but it's not available to the public yet due to risks of false positives. This move follows similar efforts by Google and Anthropic. While it's a step toward transparency, it's not a silver bullet for detecting AI text. For now, it's a regional experiment that could shape how AI content is labeled in the future.