Anthropic will begin embedding watermarks into all content generated by its Claude AI assistant, a change driven by new European Union transparency requirements.
The company confirmed Tuesday it has signed the EU AI Act’s Code of Practice on Transparency of AI-Generated Content. This framework, tied to Article 50(2), pushes AI developers to make synthetic content traceable.
Starting Aug. 2, 2026, every model Anthropic launches in the EU will ship with watermarking capabilities built in from day one. That covers output across Claude, Claude Code, Claude Cowork, Claude Tag, and the Claude Platform API. Text responses will carry embedded markers, while files will include digitally signed provenance metadata wherever platforms support it.
Meanwhile, the policy reaches far beyond Europe. Anthropic plans to roll out watermarks globally wherever Claude operates, though the company cautions that certain features or platforms may not accommodate every watermarking method.
The technical specifics remain thin for now. Anthropic says text watermarks will live inside the text itself and survive copy-and-paste operations, but offers no explanation of how detection works. For images and other files, the company will adopt the Coalition for Content Provenance and Authenticity (C2PA) open standard. Recognition details will arrive in future documentation.
Older models released before the August 2026 deadline, which includes nearly everything currently available, will also receive watermarking support retroactively.
The system has clear limits, though. Heavily edited, paraphrased, or translated text may lose its watermark. Likewise, metadata can vanish through file format conversions, screenshots, or other manipulation.
Anthropic now joins OpenAI, Google, and Meta in embracing content provenance measures.
