Anthropic has started marking Claude’s output as AI-generated, a response to Article 50 of the EU AI Act, which requires AI providers to make generated content machine-detectable. Anthropic has signed the EU’s Article 50(2) Code of Practice covering AI-generated content, and new Claude models launched from 2 August 2026 support the marking from launch; models that shipped earlier are being retrofitted during a transition period.
The mechanism differs by content type. For text, Claude weaves an imperceptible watermark directly into the generated output: a statistical pattern in the text itself that doesn’t change its meaning, quality, or readability, and that a human reader will never notice. For file types that support it, SVG, PNG, and JPG among them, Claude attaches signed provenance metadata following the C2PA (Coalition for Content Provenance and Authenticity) industry standard, the same open standard used elsewhere in the industry to record where a piece of content came from and flag tampering.
Coverage is broad by design. The marking applies across the Claude Platform API, the Claude apps, Claude Code, Cowork, and Claude Tag, and Anthropic says it will apply wherever Claude is offered worldwide, not only to users inside the EU, despite the requirement itself being European.
The company is candid about the limits. A detected mark means content may have been processed by Claude, not a guarantee of its full provenance, and the absence of a detectable mark doesn’t prove content wasn’t AI-generated: heavy editing, paraphrasing, or output from an older, pre-retrofit model can all defeat detection. The move puts Anthropic alongside other major labs that have adopted C2PA-style content credentials as regulatory and platform pressure for AI-content transparency builds globally, not just in the EU.