Anthropic implements invisible watermarks in Claude-generated text for EU AI Act compliance

Anthropic announced the implementation of imperceptible watermarks directly in text generated by Claude models, aimed at complying with the European Union's Artificial Intelligence Act. The watermarks persist even after copying and pasting, and may remain present even after moderate edits to the text. In response to concerns raised by industry experts, Anthropic provided important clarifications: the company plans to provide a free API allowing users and third parties to independently detect whether text contains Claude watermarks. Contrary to some concerns, Anthropic confirmed that watermarks do not affect the meaning, quality, or readability of Claude's responses. Watermarks indicate that Claude processed text, not proving that text was entirely AI-written, even if the model only performed revision or translation. Implementation occurs in compliance with European regulations, having started on August 2, 2026.