Photo by Jackson Sophat on Unsplash
New EU Regulation Prompts Watermarking in AI Texts
Anthropic, the US-based AI company, has introduced an invisible watermarking system for texts generated by its chatbot Claude. This move responds to newly established European Union regulations requiring that AI-generated content contains machine-readable markers. The watermarking is designed to be undetectable to users and remains intact even when the text is copied or lightly edited. This feature is already integrated into Claude versions released after August 2, 2026, with older versions planned to be updated accordingly [Source 1].
Details and Mechanism of Claude’s Watermarking
The watermarking is embedded at the model level, affecting outputs not only from the Claude app but also API connections and other services like Claude Code, Claude Cowork, and Claude Tag. It applies across multiple cloud platforms including AWS, Google Cloud, and Microsoft Foundry. The marking manifests through metadata and subtle statistical patterns within the text tokens, making it imperceptible and minimally affecting text quality. However, the watermark does not guarantee that Claude originally created the text; it may have been processed or corrected by Claude after initial human creation [Source 2][Source 5].
Anthropic has yet to release detailed technical documentation on how to verify the presence of watermarks. Developers who integrate Claude into other products will need to ensure compliance with transparency rules relevant to their services [Source 2].
Implications for Expats and Foreign Users in Germany
The introduction of visible machine-readable watermarks in AI-generated texts directly impacts expats, international students, and foreign workers using AI tools in Germany. With compliance tied to EU rules, users can expect increased transparency regarding the origin of AI-assisted content, which may affect work submissions, academic papers, and communications where AI usage disclosure might be legally required. This step can help users avoid unintended breaches of EU obligations linked to AI content disclosure.
Individuals relying on Claude or similar AI services should monitor updates and ensure their usage complies with local laws on AI content. Developers and service providers leveraging Anthropic’s technology should prepare for transparency obligations and may need to implement mechanisms to identify watermarked content in their workflows.
For more information, users can read the original report on the implementation of AI watermarking here: https://www.tagesschau.de/wirtschaft/anthropic-ki-wasserzeichen-claude-100.html [Source 1].
