What happened
Anthropic says it will watermark text produced by its models, Claude included, to comply with European rules. The company laid this out on an updated support page.
The trigger is the EU AI Act's Transparency Code, which took effect on August 2. It requires AI companies to label AI-generated or edited content so that other systems can recognize it as such.
Anthropic confirmed that every model released after August 2 will automatically ship with watermarking tech for both generated text and files. For files, it is using C2PA, an open standard.
Older models are on the list too. Anthropic says it will extend support to them.
The detail that matters most: the watermark is baked into the text. Per the support page, it applies at the model level, so it shows up whichever Claude product or surface produced the words. It travels with the text when copied and pasted, and may persist through some editing.
How much editing it takes to kill the mark is unclear. TechCrunch says it asked Anthropic to clarify.
Why it matters
This is a major AI lab turning a European transparency rule into shipping product. Where one big company goes first, the others get a template, or an awkward comparison.
Text is the hard case. A label on an image file is one thing, but words get pasted into emails, docs and homework. A mark that lives in the text itself is built for exactly that journey.
The open question is durability. "May persist through some editing" is a polite way of saying the limits are not yet known. A watermark that dies after one rewrite is a speed bump, and one that survives is a detection tool. Which one this is will decide how useful it turns out to be.
What it means for Europe
This is the AI Act doing what it said on the tin: regulation changes product behavior, and the change applies to everyone using Claude, not just people in Europe, because the mark is applied at the model level.
For European companies, schools and publishers, it means AI-written text from Claude can now be identified by other systems. Expect detection and compliance tooling to start caring about this.
For European users, it is a preview of the new normal: the Transparency Code is live, and the big labs are now working out how to meet it.
What to do with it
- Assume your Claude text is tagged.If you publish AI-assisted writing, treat it as identifiable and be upfront about it.
- Read the support page.Anthropic's updated page is the source for how the watermark behaves, and details may change as older models get covered.
- Don't count on editing to scrub it.Anthropic says the mark may persist through some editing, and the actual limit is unclear.
- Check your own vendors.If you use other AI tools in Europe, ask how they plan to meet the Transparency Code.
Ciao, sapiens. Your copy-paste habit just got a witness.


