Anthropic to add hidden watermarks in Claude AI content

In a notice published on its Claude Help Center, the San Francisco-based AI company said Claude models launched in the EU on or after Aug. 2 will support machine-readable marking at launch, with generated text carrying embedded watermarks and generated files, including digitally signed provenance metadata where applicable.

The requirement stems from Article 50 of the EU's Artificial Intelligence Act, which requires providers of generative AI systems to make synthetic content identifiable through technical means.

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

Text watermarks are designed to be invisible during normal reading and to persist when content is copied or pasted, though multiple reports on the announcement noted the company has cautioned that heavy editing can weaken or erase the signal. Files in supported formats, including common image types, are expected to 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 marking AI-generated images.

The EU AI Act's transparency requirements form part of a broader regulatory rollout in Brussels aimed at governing artificial intelligence systems by risk level, with obligations taking effect on a staggered timetable through 2026 and beyond.

As Qazinform News Agency reported earlier, Meta launched an AI coding agent to take on OpenAI and Anthropic rivals.