Claude AI Content Now Marked With Invisible Watermarks Globally

The markings will be embedded at the model level, so they will appear in output across Claude's platforms, including Claude, Claude Code, Claude Cowork, and Claude Tag, regardless of where users access the service. Anthropic said it plans to apply the system globally wherever supported Claude models are available, not just in Europe.

The text watermark is designed to be imperceptible and should not affect the meaning or readability of the content. It is expected to survive copy-pasting and some editing, but heavy rewriting, paraphrasing, translation, or combining text with other material could make it harder to detect.

For files such as SVG, PNG, and JPG images, Claude will attach digitally signed provenance metadata based on the Coalition for Content Provenance and Authenticity (C2PA) standard. This metadata can indicate that a file was processed by Claude and help determine if it was altered afterward. Availability may vary depending on the file type and platform.

Anthropic is also developing tools for users and third parties to check for these markings, and plans to publish technical documentation on how detection will work. The company cautioned that a detected mark does not prove authorship, since someone could use Claude to translate or proofread their own work, and an absence of a mark does not prove human creation, as older models, heavy editing, or short text may not produce a detectable signal.

The rollout aligns with transparency commitments under the EU's AI framework. Models launched in the EU on or after August 2 will support the system from the start, and Anthropic is working to add the capability to older models during the transition period. The company also said it will provide further technical guidance for developers building on Claude, who must assess their own obligations under the EU AI Act.