Anthropic Adds Watermarks to Claude-Generated Text Under EU Rules

Anthropic has begun watermarking text output from Claude and its other AI models to adhere to the European Union's new artificial intelligence governance framework. The watermarking system operates at the model level, meaning it applies across all Claude-based products and surfaces where users interact with the technology.

What the EU Requires#

The EU AI Act's Transparency Code came into force on August 2, 2026, and mandates that AI companies mark any content their systems generate or modify in ways that other systems can detect. Anthropic's move represents compliance with this regulatory mandate, which applies to a broad swath of the industry.

How Anthropic's Watermarking Works#

Every Claude model deployed after August 2 will automatically embed watermarking technology into both text and file outputs. For file-based content, the company is leveraging the C2PA open standard. According to the company's updated support documentation, the watermark travels with the text when copied and pasted to other locations and may survive some forms of editing. Anthropic plans to backport this functionality to older models as well.

The watermarking system covers all Claude products, including:

• Claude platform API
• Claude Code
• Claude Cowork
• Claude Tag

The company has not yet disclosed precisely how much user editing would be required to strip the watermark, leaving some uncertainty about its robustness.

A Broader Industry Pivot#

Anthropic is far from alone in this shift. Watermarking and disclosure of AI-generated content have become standard practice across the tech industry following regulatory pressure and public concern. Music platform Suno recently committed to flagging tracks created on its service, and Substack partnered with Pangram to identify AI-assisted newsletters. Other major players—including Black Forest Labs, Google, Meta, Microsoft, OpenAI, and Synthesia—have similarly signed onto the EU transparency code.