Anthropic has introduced watermarking for text and provenance metadata for media generated by its Claude AI models, effective worldwide. The watermark is embedded within text responses and travels with the text when copied or edited. Media files like .svg, .png, and .jpg now include metadata following the Coalition for Content Provenance and Authenticity (C2PA) open standard. Claude models launched in the EU on or after August 2, 2026, support machine-readable marking at launch, according to medianama.com.
The watermarking applies at the model level, ensuring it is present across all Claude products and surfaces. The provenance metadata contains signed labels that indicate a file was processed by Claude and helps detect tampering. Anthropic clarified that these measures are implemented worldwide and are part of compliance with the EU AI Act. The company plans to release detection tools and detailed technical documentation soon, although these are not yet available to users or third parties.
This move aligns with broader industry efforts to increase transparency and authenticity of AI-generated content. The use of C2PA metadata is becoming a standard for content provenance across AI and media companies. By embedding watermarks and metadata, Anthropic aims to address concerns about misinformation and manipulation of AI outputs. The EU AI Act mandates such transparency measures, making Anthropic one of the early adopters of machine-readable marking in AI models within the region.
Anthropic’s implementation of watermarking and provenance metadata for Claude models worldwide marks a key step in AI content accountability. The company confirmed that Claude models launched in the EU from August 2, 2026, support machine-readable marking, with global rollout already in effect. Detailed technical documentation and detection tools are expected to be published soon, providing users and developers with resources to verify AI-generated content authenticity.