Anthropic will now add machine-readable marks to content generated by its Claude AI models, including embedded watermarks in text, to help identify content that has been generated or processed by Claude.
The update is part of the company’s commitments under the European Union AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content. The markings are intended to provide users and third parties with information about the origin and processing of AI-generated content.
The company said new Claude models launched in the EU on or after August 2, 2026, will support machine-readable marking from launch. Generated text will carry embedded watermarks, while supported files will include digitally signed provenance metadata.
The markings will apply to supported Claude models across the company’s API, Claude, Claude Code, Claude Cowork and Claude Tag products, as well as when Claude is offered through cloud partners such as AWS, Google Cloud and Microsoft Foundry. Anthropic said some platforms and features may not support every type of marking.
“New models will mark AI-generated content from day one,” the company said.
Anthropic is also working to add marking support to models released before August 2, 2026, which are covered by a transition period under the EU AI Act.
How Claude will mark content
Anthropic said Claude will use two methods: embedded watermarks in text and signed provenance metadata for supported files.
For text, Claude will embed an “imperceptible watermark” directly into generated content. The watermark is designed to remain with the text when it is copied and pasted and may persist through some editing.
For files such as SVG, PNG, and JPG, Claude will attach signed provenance metadata based on the Coalition for Content Provenance and Authenticity (C2PA) open standard. The metadata can indicate that a file was processed by Claude and help determine whether it has been altered.
Anthropic is also developing tools that will allow users and third parties to detect Claude’s watermarks and provenance metadata. This might give rise to new businesses, which will work towards identifying and removing the AI watermarks and humanise the content.
Limits of the marking system
The marks are intended to provide signals about whether content was processed by Claude but are not conclusive proof of its origin.
For example, content may carry a Claude mark even when the original ideas, text or data came from another source and Claude was used only for tasks such as proofreading, translation or summarisation. Marked content can also be changed, shortened or combined with other material after being processed by Claude.
The absence of a detectable mark does not necessarily mean that content was not generated or processed by AI. Marks may be lost if text is heavily edited, paraphrased, translated or combined with other writing. Very short passages may also not contain enough text for reliable detection.
For files, metadata can be removed through processes such as format conversion, re-saving or taking screenshots. Marks may also be unavailable when a platform, feature or file type does not support a particular marking method.
Anthropic said developers using Claude in their own products should independently assess the transparency requirements that apply to their services under Article 50 of the EU AI Act.