Anthropic has confirmed that new Claude models will carry an invisible signal inside the text they generate. The move responds to transparency rules under the EU AI Act’s Article 50, and it takes effect for any Claude model launched on or after August 2, 2026.
Anthropic says the marking applies worldwide, not just to European users, covering the API, the chat app, Claude Code, Claude Cowork, and Claude Tag, including access through AWS, Google Cloud, and Microsoft Foundry.
This is not a visible tag or a sentence declaring that a piece of writing came from AI. It works at a statistical level instead. When Claude generates text, it picks each word from a range of plausible options, and the watermarking system nudges those choices according to a hidden pattern.
The writing still reads naturally, but a tool built to look for that pattern can detect it. Since the signal lives inside the sequence of words rather than in a file’s metadata, it should survive ordinary copying and pasting, and Anthropic says it may hold up through some light editing too.
READ: Kimi K3 Launches as China’s Largest Open AI Model to Rival OpenAI and Anthropic
The company has not yet published the underlying algorithm or released a public detector, so how well the policy works in practice is still unclear.
The bigger catch is what a detected watermark actually proves. It only shows that a supported Claude model processed the text somewhere along the way, not who came up with the ideas.
Someone could write an entire essay themselves and ask Claude only to fix grammar or translate a paragraph, and the resulting text could still carry the mark. The opposite is just as true, no detectable watermark does not mean a human wrote something.
Detection can fail after heavy rewriting, translation, mixing AI text with human text, or when a passage is too short to carry enough signal.
Older Claude models released before the August cutoff will not feature this until Anthropic rolls it out during a transition period, and no firm date has been set for that.
Files will be handled differently from text. For supported formats such as PNG, JPG, and SVG, Anthropic is adding digitally signed provenance metadata based on the C2PA standard. This is an open framework used across the industry to track how digital files were created or edited.
Unlike the text watermark, this metadata is stored separately from the file’s actual content. That means it can be removed by taking a screenshot, converting the file to another format, or simply saving it again.
Anthropic frames all of these measures as a transparency measure, not a definitive authorship check, and says detection tools for outside parties are still being built. For now, the safest way to read a Claude watermark is as a clue about where text passed through, not a verdict on who wrote it.
