Claude's AI Models to Mark Generated Content for EU Compliance Starting 2026

Claude's AI models will embed invisible watermarks and associate metadata with AI-generated content as part of compliance with the EU AI Act. This initiative begins on August 2, 2026, marking content across the Claude Platform and global operations, ensuring transparency and adherence to regulatory measures.

Claude is set to introduce a groundbreaking feature in its AI models to mark AI-generated content, aligning with the European Union's AI Act's Article 50(2) requirements on transparency. This initiative will commence on August 2, 2026, applying to AI models launched in the EU from that date onward.

The company plans to utilize two complementary techniques: embedding invisible watermarks directly into text and attaching signed provenance metadata to files. These watermarks will maintain the quality and readability of the content, allowing them to stay with the text even when copied. For file formats like .svg, .png, or .jpg, the metadata will comply with the Coalition for Content Provenance and Authenticity (C2PA) standard.

The implementation will cover various platforms and regions globally, ensuring that AI-generated content can be identified regardless of where Claude models are used. The company acknowledges a transitional phase for models released prior to the deadline, with plans to extend marking support to them. Furthermore, Claude's systems will have distinct marking mechanisms tailored to cloud distribution channels' capabilities.