Claude AI to Mark Content: EU Transparency Rules Explained

Claude models will soon mark AI-generated content. This means a digital trail will be embedded that machines can read. Anthropic is implementing this as part of its commitment to the EU AI Act's transparency rules. Claude models launched in the EU on or after August 2, 2026, will support these machine-readable marks. The system uses two methods: an invisible watermark in text and signed provenance metadata for supported files like images. The text watermark is embedded directly without changing how it looks. It's designed to stay with the text when copied, helping detection tools identify content that has passed through Claude. For image files, metadata based on the C2PA open standard will be attached. These marks will apply across various Claude platforms globally, though platform support for each marking type may vary. There is also a transition period for older Claude models. It's important to note that a Claude mark shows content was processed by the system, not that Claude is the sole author. Also, the absence of a mark doesn't mean AI wasn't involved. The goal is transparency, offering another way to establish if content passed through Claude. This helps users understand the origin of digital information.