Does Claude watermark its text?
Yes. Claude models launched on or after 2 August 2026 carry a SynthID-Text watermark, and no tool can remove it, including ours.
Anthropic published how it works. Instead of an ordinary random number choosing between equally good next words, a key and the preceding words choose, and the pattern of those choices is the watermark. Their words are: nothing is added to the text and there are no hidden characters. It is applied globally, driven by the EU Code of Practice on Transparency of AI-Generated Content, and a detection API has been announced but not released. Light editing probably will not remove it; a rewrite where every word changes will.
What Clipboard Sanitizer can do about it
Nothing about the watermark, and we will not pretend otherwise. What we can do is clean the ordinary characters that come with any pasted text: curly quotes, em dashes, exotic spaces, and whatever invisible characters made it through. That is a real problem and a different one. If you came here to remove Anthropic's watermark, no product can do that, and the ones claiming to are selling you nothing.
Questions
Will removing invisible characters remove Claude's watermark?
No. The watermark is in the choice of words, not in any character. Removing a zero-width space or straightening a quote does not touch it.
Does the watermark identify me?
Anthropic says not: it identifies that Claude produced or processed the text, and carries nothing about the user, their organisation or their conversations.
Does every Claude reply carry it?
It rides on entropy, so creative writing carries a strong signal and short factual answers or code carry a weak one. Anthropic describes it as sparse in factual text and code.