Invisible, machine-readable watermarks are now embedded in text from new Claude models. The update meets EU transparency rules but will be active worldwide. Detection tools and documentation are coming soon, with key limitations for authorship verification.
Anthropic has begun embedding invisible, machine-readable watermarks into text generated by its latest Claude AI models, expanding the feature globally to meet new transparency requirements under the European Union’s AI Act. The company confirmed that watermarking will be active for all users, not just those in Europe, and will apply to Claude models launched in the EU on or after August 2.
The watermarking system covers output from supported Claude products, including Claude, the Claude API, Claude Code, Claude Cowork, and Claude Tag. Anthropic is also working to extend watermarking to models released before August 2 during the AI Act’s transition period. The company says the watermark is imperceptible to users and does not affect the meaning, quality, or readability of the generated text.
Because the watermark is embedded directly in the text, it remains present when content is copied, pasted, or lightly edited. Watermarking is implemented at the model level, ensuring that all supported Claude products include the mark in their outputs. For images, supported PNG, JPG, and SVG files will receive signed C2PA provenance metadata to indicate AI involvement.
Anthropic plans to release detection tools and technical documentation to help users and third parties identify watermarked text and signed metadata. However, the company cautions that detection has limits: a watermark only indicates that content may have been processed by Claude, not that Claude originated the ideas or wrote the original text. The absence of a watermark does not guarantee content is human-written, as older models, heavily edited text, or files with stripped metadata may lack detectable marks.
Transparency is the primary goal of this update, as Claude is widely used for creating and refining marketing and editorial content. The move could influence how AI-generated material is identified and disclosed across industries. For context on how AI features are surfacing in other platforms, see how Google is experimenting with new AI-powered buttons on its search homepage in this related report.
Anthropic has published a support page detailing how Claude marks AI-generated content, with further resources and detection tools expected soon.
