AIBRIEF
Anthropic's image watermark uses a metadata standard already adopted by several major AI companies; the text watermark is opaque, with Anthropic declining to name the method or verify that it survives editing.
Anthropic says it will start marking text and images produced by Claude with machine-readable watermarks and provenance metadata, in response to new EU AI Act transparency obligations that took effect on August 2.
The image side uses C2PA, a metadata standard for content provenance that several major AI companies have adopted. Files generated by Claude will carry a signed stamp platforms and readers can inspect to confirm an AI origin. The text side is murkier. Anthropic calls it an "imperceptible watermark" applied at the model level but does not name the method. The company says the mark survives copy and paste and "may persist" through light editing, though those claims are not independently verified in the public source.
New Claude models will mark AI-generated content from day one upon release. Support for existing models is a work in progress during a four-month compliance grace period the EU has set for products launched before the August 2 deadline. The marking will extend across Claude Platform, the Claude assistant, Claude Code, Claude Cowork, and Claude Tag.
The Verge first reported the move. The next milestone is whether Anthropic or outside researchers can show the text watermark survives paraphrase, translation, or modest edits. That persistence is what would make the marking useful. Anthropic's support documentation describes how Claude-generated content is marked today.