Anthropic confirmed on August 11 that Claude models launched on or after August 2, 2026 now embed an imperceptible statistical watermark directly into generated text, and apply the marking worldwide rather than only for users in Europe. The system covers the Claude Platform API, claude.ai, Claude Code, Claude Cowork, Claude Tag, and Claude models accessed through AWS, Google Cloud, and Microsoft Foundry. Files such as images receive a separate technique: cryptographically signed provenance metadata following the C2PA open standard.
The immediate trigger is Article 50 of the EU AI Act, which became enforceable on August 2, 2026 and requires generative AI providers to embed machine-readable marks in synthetic content so downstream platforms and regulators can identify it. Anthropic signed the EU's Code of Practice on Transparency of AI-Generated Content and says the text watermark travels with content when copied and pasted, and may persist through some editing, though older models fall under a transition period running until December 2026.
Anthropic was explicit about the system's limits: a detected watermark does not prove Claude authored a piece of text, since people commonly use the model to edit, translate, or summarize their own writing. Conversely, an absent watermark does not prove content is human-made, since heavy editing, paraphrasing, translation, short passage length, or file-format conversion can all strip the marking. A public detection tool has not yet shipped. The move follows reporting that OpenAI's own EU compliance documentation had described text watermarking as harder to deploy at scale, making Anthropic the first major frontier lab to roll out production watermarking across its full product line at once.
