Anthropic Claude Watermark

Claude-generated text now carries an invisible watermark. Not just in Europe. Everywhere.

Anthropic announced this week that it's embedding a machine-readable mark into text from current Claude models, plus signed provenance data on files using the same C2PA standard OpenAI, Google, and Adobe use. The mark is designed to survive copy-paste and hold up through some editing.

The EU AI Act's transparency rule is the legal trigger. But Anthropic chose to roll it out worldwide instead of just where the law applies, and reaction online has been mixed. Some people see it as responsible disclosure. Others see it as a company quietly fingerprinting everything you write with it.

Anthropic's own caveat is the part I keep coming back to: this doesn't prove who wrote something. If you use Claude to edit or translate your own work, that text can still carry the mark. Provenance and authorship aren't the same thing, and I think a lot of people are going to conflate them.

Would knowing a piece of writing passed through an AI model change how you read it, even if a person wrote most of it?