Anthropic has started embedding invisible watermarks into text generated by its Claude chatbot, a move the company says is necessary to comply with the EU AI Act’s Transparency Code, which now requires tech companies to label AI-generated or AI-edited content.
The watermarking system inserts machine-readable labels directly into Claude’s output text. According to Anthropic, the watermarks are designed to not alter the meaning, quality, or readability of the generated content. Because the watermark is embedded within the text itself, it can survive copy-paste operations and some editing, a feature that has proven controversial among users.
The backlash has been swift. Users on Reddit and other social media platforms have complained that the watermarks will effectively expose anyone using Claude at their jobs or in school. One widely shared Reddit post described the system as a digital tattoo, arguing that students who ask Claude to reorganize a paragraph or writers seeking synonyms will be caught.
Anthropic’s watermarking effort applies to new Claude models launched in the European Union on or after August 2, 2026, and the company says the approach will be applied worldwide. The watermarks work across all access points, including Claude applications, the API, and services offered through cloud partners such as AWS, Google Cloud, and Microsoft Foundry.
The company has acknowledged that a detected watermark does not provide conclusive proof that content was entirely generated by AI. A detected label only indicates that Claude may have processed the content, leaving room for partial human involvement. Older Claude models released before August 2, 2026 will receive watermarking support in the future, though no specific timeframe has been announced.
The decision places Anthropic at the forefront of AI transparency compliance. While other companies including OpenAI and Google have pledged to research watermarking technologies, Anthropic appears to be the first major AI lab to deploy invisible text watermarks at scale. Google has open-sourced its SynthID watermarking tool, but it is primarily focused on images, audio, and video rather than text.
The controversy highlights a growing tension in the AI industry between transparency mandates and user expectations. As governments worldwide tighten regulations around AI-generated content, companies face the challenge of balancing compliance with maintaining user trust and adoption. Anthropic’s experience may set a precedent for how the industry navigates this balance.
Sources: TechCrunch – Claude users mad about watermarks, Trending Topics – Anthropic watermark plans
discussion