Anthropic to embed watermarks in Claude AI under EU rules

Anthropic announced this week it will begin embedding hidden digital watermarks into text and files produced by its Claude artificial intelligence models, a move the San Francisco-based company tied directly to new transparency obligations under the European Union's Artificial Intelligence Act.

EU launch requirements

The company said Claude models launched in the EU on or after Aug. 2 will support machine-readable marking at launch. Generated text will carry embedded watermarks, while files will include digitally signed provenance metadata where supported.

The obligation stems from Article 50 of the EU's AI Act, which requires providers of generative AI systems to make synthetic content identifiable through technical means. Anthropic noted the implementation aligns with the regulation's phased rollout timeline.

Global application

Coverage of the new markings extends beyond Europe. Anthropic stated the markings cover output from supported models everywhere Claude is used, including its API-based Claude Platform, consumer chat product, Claude Code, Claude Cowork, and Claude Tag.

The deployment marks one of the first major implementations of the EU law's technical requirements by a leading AI developer. Industry observers said the move could establish a baseline for how other American technology firms approach the Brussels regulation.

Technical implementation

Text watermarks are designed to remain invisible during normal reading and to persist when content is copied or pasted. However, Anthropic noted that heavy editing can weaken or erase the signal.

Files in supported formats, including common image types, will carry signed provenance metadata built on the Coalition for Content Provenance and Authenticity, or C2PA, standard — an industry framework already used by other technology firms, including Google, for labeling AI-generated images.

Regulatory timeline

The EU AI Act's transparency requirements form part of a broader regulatory framework in Brussels aimed at governing artificial intelligence systems by risk level. Obligations under the landmark legislation are phasing in on a staggered timetable through 2026 and beyond.

The law represents the world's first comprehensive attempt to regulate generative AI through binding technical standards. Additional provisions targeting high-risk applications are scheduled to take effect in subsequent phases of the rollout.