Anthropic explains how Claude’s invisible text watermarks will work

Anthropic has outlined its strategy for embedding invisible watermarks into text generated by its Claude artificial intelligence models, a move designed to comply with emerging transparency regulations in Europe. Announced on Friday, the initiative addresses the European Union’s AI Act, which mandates that synthetic content across multiple media formats include machine-readable identifiers. The watermarking system will operate alongside Anthropic’s existing C2PA compliance for Claude-processed images, establishing a unified approach to content provenance that aligns with the bloc’s digital transparency standards.

The text marking technology utilizes SynthID-Text, an open-source watermarking framework originally developed by Google DeepMind. Rather than altering the visible output, the system embeds detect