Claude Gets Invisible AI Watermarks Starting Now

Anthropic is adding invisible watermarks to text generated by future Claude models, complying with the EU AI Act requirement that took effect August 2. The watermarking uses a cryptographic key to subtly influence word choices during text generation, leaving a detectable pattern without altering meaning or quality. The watermark carries no personal identifying information and cannot be traced to specific users or organizations. Internal testing and Google DeepMind research on the same technique found no measurable impact on output quality, creativity, or readability. Other major AI providers are implementing similar watermarks under the same EU Code of Practice.