Claude Watermark: ChatGPT Follows Suit, but Only in the EU
Basic Tutorials
Picture: Basic Tutorials
Since August, Anthropic has been adding an invisible watermark to all text generated by new Claude models, and OpenAI is now following suit with ChatGPT. The Claude watermark is embedded in the wording itself: You can’t see it, it remains intact when copied, and it applies worldwide. OpenAI, on the other hand, is currently only marking ChatGPT texts within the EU. The catalyst for both is the EU AI Act, which requires machine-readable labeling of AI-generated content as of August 2, 2026. How does the Claude watermark work? A language model selects from several suitable candidates for each word. This is exactly where the watermark comes in: A secret key slightly favors certain variants among equivalent formulations.
This report was first published by Basic Tutorials. What you see here is a brief excerpt. The full text is theirs.
Continue reading at Basic Tutorials
