Anthropic to Watermark Claude Text and Images Under EU AI Act
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Anthropic pledged to mark Claude-generated text with embedded watermarks and images with C2PA provenance metadata, complying with the EU AI Act's August 2 transparency requirements.
- New Claude models will carry the marks from day one upon release, while support for existing models is a work in progress during the EU's four-month compliance grace period for products launched before August 2.
- For images, Anthropic will apply C2PA — a provenance metadata standard already embraced by Adobe, OpenAI, and Google — but the company did not name the system used to imperceptibly watermark text output.
- The text watermark is woven directly into Claude's responses without changing meaning, quality, or readability, and is designed to travel with copied-and-pasted text, potentially persisting through some editing.
- Watermarking will apply globally across Claude Platform (API), Claude, Claude Code, Claude Cowork, and Claude Tag — including when Claude models are accessed through AWS, Google Cloud, or Microsoft Foundry.
- Anthropic is developing third-party detection tools with technical documentation to follow, though it's unclear whether existing detectors like Google's Gemini chatbot will read C2PA metadata in Claude-generated files.
- The company itself hedged that the marking systems are "far from infallible," and C2PA data is known to be easily stripped — sometimes accidentally during platform uploads.
Why it matters: Anthropic becomes one of the first major AI labs to formally commit to dual watermarking (text + image) under the EU AI Act, but its own caveats about C2PA's fragility and the unnamed text watermarking system mean provenance metadata still can't reliably flag AI-generated content. The four-month grace period gives existing Claude products runway before enforcement.
Ask SkimNews
