Anthropic will embed invisible watermarks into all text and files generated by its Claude artificial intelligence models, a move the company says is designed to comply with new European Union transparency regulations.