To comply with the EU AI Act: Anthropic announces that new Claude models will embed invisible watermarks in text, globally activated. To meet the EU AI Act’s transparency requirements for generative content, Anthropic has announced that starting August 2, 2026, all new Claude AI models will embed invisible watermarks within generated text—undetectable to the human eye. This mechanism will be enforced globally, not just in Europe, covering all channels including APIs, https://t.co/Lh0Ycot08v, Claude Code, Claude Cowork, and third-party cloud platforms such as AWS, Google Cloud, and Microsoft services. According to reports, the watermark is implemented at the model level: machine-readable signals are embedded directly into the token sequences of generated text. The watermark does not affect semantics, quality, or reading experience, but can be detected using specialized tools to determine whether content was generated by Claude. Because the watermark is embedded directly into the text itself, it may persist even if users copy, paste, rewrite, or make moderate edits. For images and vector files, the system will also attach digital signature metadata compliant with the C2PA standard to enable traceability. #Anthropic #Claude #EUAIAct #InvisibleWatermark #AIGeneratedContent #C2PA #Transparency #AIRegulation #TechNews