Anthropic’s Claude will watermark AI-generated text. Here’s how it works

Anthropic, the creator of the Claude chatbot, has announced it will begin watermarking AI-generated text and files to ensure transparency and comply with the European Union's AI Act. The EU's AI Act, which came into effect on August 2nd, necessitates transparency and labeling of AI-generated content. Claude models launched on or after August 2nd will contain an imperceptible watermark woven into the text through backend coding, which will travel with the text when copied and pasted or edited.

For AI-generated files like JPG images, Anthropic will incorporate signed provenance metadata to indicate the file's origin and any tampering. While the watermarking system provides a level of transparency, it is not definitive proof that content is AI-generated. For instance, text entered into Claude for editing would still carry the watermark.

Companies are quickly adopting AI transparency measures as AI-generated content becomes more prevalent. Spotify has also pledged to label AI-generated artist personas and not include them in personalized music recommendations, while encouraging artists to disclose if AI components were used in song creation. The EU's AI Act requires companies to implement watermarking or identifying code or face substantial fines.