Anthropic explains how Claude's invisible text watermarks will work

It's an open-source technology by Google DeepMind that embeds invisible watermarks in AI-generated text using word probability patterns.

Text quality remains unaffected, but watermarks enable content provenance verification, aiding regional institutions in meeting international standards.

Yes under the EU AI Act for European markets, but voluntary elsewhere, though it may become an industry norm.

They are designed to be tamper-resistant, but effectiveness varies; some limitations may exist.