Anthropic to watermark Claude text and images worldwide

Anthropic will begin embedding invisible, machine-readable watermarks into text and images generated by its Claude models, a move designed to comply with the European Union's AI Act transparency rules. The company announced the change in a support document, stating that new Claude models launched on or after August 2, 2026 will support marking at launch, with support for existing models to follow. The marking will apply globally, not just in the EU, across all Claude products including the API, Claude Code, and Claude Cowork Source: The Verge, Euronews.

How it works

For text, Claude will weave an imperceptible watermark directly into the generated words, which does not alter meaning or readability and travels with the text when copied or lightly edited Source: Gizmodo. For files, such as .svg, .png, and .jpg, Claude will attach signed provenance metadata using the C2PA open standard, which can signal whether a file has been tampered with Source: Tom's Hardware. Anthropic says it will provide detection tools and share details in forthcoming technical documentation Source: TechCrunch.

Caveats and limitations

Anthropic is explicit that the system is not foolproof. A detected watermark only indicates content "may have been processed by Claude," and the absence of a mark does not rule out AI generation, as heavily edited or paraphrased text may lose the signal Source: The Next Web. C2PA metadata can be stripped through format conversion or screenshots Source: Euronews. Tom's Hardware notes that Anthropic made no mention of actual image watermarking, despite it being a requirement of the Code of Practice, and expects the firm to implement it later Source: Tom's Hardware.

Context

Anthropic joins OpenAI and Google in outlining compliance with the EU AI Act, which took effect on August 2, 2026. Non-compliance can trigger fines of up to €15 million or 3% of global annual turnover Source: Euronews. The company's global rollout highlights how EU standards are influencing AI practices worldwide Source: Euronews.