Anthropic Adds Invisible Watermarks to Claude AI Outputs for EU Compliance

Anthropic has begun embedding invisible watermarks into text and files generated by its Claude artificial intelligence models, a global implementation designed to comply with new European Union transparency regulations. The company announced the change in an updated support page on Tuesday, August 11, 2026, stating that the watermarking applies to all models launched on or after August 2, 2026, with plans to extend the feature to older versions in future updates.

The move is a direct response to the EU AI Act's Article 50, which took effect on August 2, 2026. This regulation requires providers of AI systems that create synthetic content to mark their outputs as artificially generated. Non-compliance can result in significant fines. Anthropic has signed the associated Code of Practice on Transparency of AI-Generated Content, joining other major technology firms like Google, Meta, and Microsoft in adhering to the new standard.

According to Anthropic, the watermarking is applied at the model level, meaning it will be present in output from all Claude products and services. This includes the consumer Claude app, the developer-facing Claude Platform API, Claude Code, Claude Cowork,...