Anthropic: future Claude models will carry an invisible watermark, with no name
On 14 August 2026, Anthropic explains that future Claude models will mark their text. The watermark would not add hidden characters, would not cost more, and would not, according to the page, carry a way back to a person, an organization or a chat.
Getting started
Original AI-generated illustration · SecuFocus
At a glance
Key points
This is not a setting you see. It is a mark in the choice of words, to estimate whether Claude wrote the text. The page ties it to the European duty, from 2 August 2026, to mark generated content. It does not say that models already in service do this.
What changes
The page is dated 14 August 2026. It speaks of coming models, not of a switch on Claude today. The chosen method, Anthropic says, does not in practice change quality or content, is not visible to a reader, adds nothing to the text, and does not ask for extra tokens.
The watermark would not be specific to Claude. Other publishers who signed the same code of practice must mark their outputs for the European Union market. Recognizing “AI” is therefore not recognizing Claude.
What leaves
Anthropic says the mark carries no identifying information, and that it cannot be traced to a person, an organization or a chat. If that sentence holds, the watermark is not an account tracker. It remains a statistical signal about where the text came from.
It does not encrypt the conversation, and it does not stop the conversation from leaving for Anthropic. Marking the output and keeping the input are two different gestures.
What we did not verify
We did not submit a text to a detector, or check that a current Claude is still unmarked. “Future models” is the page’s phrase.
The decision
Do not count on this watermark to know who wrote a page, or to hide that you used Claude. It does not identify the author. It also does not replace rereading what you send.
Check and explore
Sources for this article
Numbers connect each reference to the passages that use it. Dates show when the documentation was consulted.
- Claude text watermark, Anthropic ↗anthropic.com ·
This article draws on the sources above. The exercises are for you to try on your devices; SecuFocus does not present them as tests carried out by its editorial team. Interfaces and features can change. Method and corrections.
Cite this article
Keep this reference with the article when you save or share it.