Anthropic has begun embedding invisible watermarks into text and files generated by its Claude artificial intelligence models, a global implementation designed to comply with new European Union transparency regulations.