Anthropic will add invisible, machine-readable watermarks to all Claude AI outputs from models released after 2 August 2026, and extend the system to older models by 2 December 2026 worldwide. The move aims to comply with Article 50 of the EU AI Act, which requires synthetic audio, image, video, or text to be machine-detectable as artificially generated, supporting trustworthy AI and protection of fundamental rights. Watermarks may appear in content or metadata and can sometimes be removed or absent, limiting reliability. The rollout could reshape legal and professional practices around confidentiality, accountability, provenance, and global AI transparency standards.
This update represents a notable development in the Ai sector. Organizations and founders tracking this space should evaluate potential strategic and technical implications on their operations.