Anthropic adds invisible watermarks to Claude texts for compliance

Anthropic has introduced an invisible watermarking system for texts produced by its Claude AI models, ensuring content remains identifiable even after copying, pasting, or modifications. This innovation is designed to meet the European AI Act's transparency requirements, which became enforceable on August 2, 2026. Anthropic is also developing detection tools for third parties, though the watermark signal may weaken after extensive text transformations.

The watermarking system embeds metadata into the text output without altering its readability, making it a subtle yet effective tool for tracking AI-generated content. While the technology is a step toward combating misinformation and ensuring compliance with regulatory standards, its robustness against adversarial attacks remains a challenge. Anthropic’s approach reflects a broader industry trend toward responsible AI deployment, particularly as governments worldwide tighten regulations on generative AI tools.