Anthropic to Add Watermarks Identifying AI-Generated Text
Anthropic announced that future AI models will embed invisible watermarks in generated text to identify AI-created content. The move aligns with new EU AI transparency laws and will be applied globally.
AI company Anthropic stated on Tuesday that its forthcoming AI models will embed an invisible watermark within all generated text, identifying it as AI-created.
The change is a response to new European Union laws regarding AI transparency, though Anthropic indicated it will apply in all countries where its Claude models are utilized. Currently, no U.S. law mandates such disclosures for AI-generated content.
Anthropic has signed the EU AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content, which includes watermarking provisions. The company has outlined its plan for implementing these requirements.
Claude models released after August 2, 2026, will begin generating embedded watermarks in text at launch. Under EU law, AI providers have until December 2 to upgrade existing models to support the machine-readable marking requirement.
While specific technical details of the watermarking process remain undisclosed, Anthropic assures that the watermark will "travel with the text" when copied and pasted. The company also plans to inject standard provenance data into generated images, video, and audio to meet California's transparency law requirements.