The Brief:
Anthropic is adding an invisible watermark to everything Claude writes, to comply with the EU AI Act’s new transparency rules.
Legal AI runs on models like Claude, so the mark could end up embedded in contracts, memos and other work product produced with AI help.
Anthropic dropped a big update this week. Claude will start weaving an invisible watermark into everything it generates.
The move complies with the EU AI Act, which requires AI providers to make their outputs detectable as AI-generated. The law only applies in the EU, but Anthropic’s marking Claude’s outputs globally anyway.
The mark is built to survive real-world use. “Because the watermark is part of the text, it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing,” the firm stated. “Watermarking will be applied at the model level, which means it will be present no matter which Claude product or surface the text comes from.”
Anthropic also said it will release detection tools, letting users and third parties check whether a piece of text or file carries a Claude watermark.
A detected watermark doesn’t mean Claude wrote the document. It means Claude touched it. Someone who used Claude to proofread, translate or summarise their own work could still produce a marked file, even though the underlying content is entirely theirs. The mark proves involvement, not authorship.
Legal AI platforms like Harvey and Legora run on top of models like Claude, so a clause drafted with AI help and pasted into Word could carry a trace of Claude’s involvement.
Does it actually change anything? Maybe.
Clients are already pushing hard on this front. Research from Integris found 85% of law firm clients want their law firm to disclose AI use, with nearly half calling it “extremely important.” Right now, that disclosure relies almost entirely on firms volunteering the information.
Firms can no longer count on AI use staying behind closed doors. Now the documents can do the talking.