Anthropic says Claude watermarks introduced under EU rules will apply worldwide

Anthropic will embed invisible watermarks in text generated by new Claude models as the company moves to comply with transparency requirements under the European Union’s AI Act.

Claude models launched in the EU on or after August 2, 2026 will support machine readable marking from launch, according to an updated Claude support document detailing how Anthropic plans to implement its commitments under the EU AI Act’s Code of Practice on Transparency of AI Generated Content.

Generated text will carry embedded watermarks, while supported files such as PNG, JPG and SVG files will include digitally signed provenance metadata using the C2PA open standard.

While the requirement stems from European regulation, Anthropic said the marking will not be limited to the EU. Supported models will mark their output wherever Claude is offered worldwide.

The text watermark will be embedded directly into Claude’s output and will not be visible to users or change the meaning, quality or readability of the response. 

Because the watermark is embedded in the text itself, Anthropic said it will travel with the text when users copy and paste it elsewhere and may remain detectable after some editing.

However, heavily editing, paraphrasing, translating or combining the text with other material could make the watermark undetectable. Very short passages may also contain too little information for reliable detection.

The marking system will apply at the model level across supported Claude products, including Claude Platform API, Claude, Claude Code, Claude Cowork and Claude Tag.

Anthropic also said embedded watermarks will apply when supported Claude models are accessed through AWS, Google Cloud or Microsoft Foundry, although signed provenance metadata may not be available through every platform.

The company is working to add marking support to Claude models released before August 2. It also plans to provide tools allowing users and third parties to detect Claude watermarks and provenance metadata, with further technical documentation expected later.

Anthropic cautioned that detecting a Claude mark does not conclusively prove Claude originally authored the content. Claude may have instead proofread, translated, summarized or otherwise processed material that originated elsewhere.

Likewise, the absence of a detectable mark does not prove content was not generated or processed by Claude, as marks can be lost through heavy editing, translation, file conversion or other modifications.

The changes come as transparency requirements under Article 50 of the EU AI Act take effect. Anthropic has signed the EU AI Act’s Code of Practice on Transparency of AI Generated Content alongside companies including Black Forest Labs, Google, Meta, Microsoft, OpenAI and Synthesia.