🔥 Claude’s invisible AI watermark is starting to make a lot more sense.

Anthropic says Claude will use SynthID-Text, which subtly influences word choices to create a statistical pattern that can later be detected with the right key.

Anthropic also plans to release a watermark detection API, while code should contain significantly less watermarking because Claude has fewer arbitrary choices when producing functional code.

Anthropic says you won’t see it, normal editing probably won’t erase it, and it shouldn’t change the quality of Claude’s writing.

Here’s the real story:

This isn’t a hidden tag buried inside a document. The fingerprint is effectively woven into the language Claude generates.

Heavy rewriting can destroy it, while light editing may leave enough of the pattern behind to identify the text.

And Anthropic says other major AI companies that signed the EU’s Code of Practice are planning watermarks of their own.

Invisible AI provenance could be about to become standard.