To meet European Union regulatory standards, Anthropic has stated that its Claude models will include identity markers to support labeling of AI-generated material. The company has joined the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content. This means Claude models released in the EU from August 2 onward will feature machine-readable markings from launch. Earlier versions will follow a transition schedule. Anthropic noted that generated text will contain embedded watermarks, while supported files will carry digitally signed provenance metadata. The approach targets text in particular, as it can be harder to distinguish from human writing compared to images or artwork. The watermark is designed to remain invisible to readers and to persist through copying, pasting, and minor edits. For file types such as .svg, .png, and .jpg, signed metadata will be attached. Limitations exist, including cases where marked content may combine human and AI input or where detection fails due to heavy editing, brief passages, or removed metadata. The markings will cover supported Claude models on platforms including the API, Claude, Claude Code, Claude Cowork, and Claude Tag worldwide, though not every feature will support every type. Access via AWS, Google Cloud, or Microsoft Foundry will still apply the watermarks. Anthropic plans to release further details on detection tools for users and third parties. The company emphasized that such signals can provide context as AI content grows more common and help meet legal requirements.
