Anthropic implements AI text watermarking to meet EU regulations

Anthropic has begun implementing machine-readable watermarks into the text generated by its Claude AI models. This move follows the implementation of Article 50 of the EU AI Act, which mandates that providers of synthetic content make their outputs machine-detectable. The regulation applies to providers serving the EU market, with potential penalties reaching €15 million or 3% of global annual turnover.

Unlike visual watermarks, text watermarking operates at the token level during the word selection process within the model. This method is invisible to human readers and does not involve hidden characters, but it remains statistically detectable using a matching cryptographic key.

In July 2026, approximately 190 organizations, including Google, OpenAI, Meta, and Microsoft, signed the EU's Code of Practice on Transparency of AI-Generated Content. While some systems already on the market are granted a transition period, providers must bring existing models into compliance by December 2, 2026. Google has already utilized SynthID-Text within Gemini since 2024.