Anthropic implements invisible watermarking for Claude AI text

Anthropic has implemented an invisible, machine-readable watermark for text generated by its Claude AI models. This feature applies to all models launched on or after August 2 and is being deployed globally across various platforms, including the chatbot, API, Claude Code, and cloud deployments via AWS, Google Cloud, and Microsoft Foundry.

Utilizing Google DeepMind’s SynthID-Text technology, the system creates a statistical signature by slightly adjusting token probabilities during generation. This signature is designed to travel with the text even when it is copied, pasted, or integrated into code. Anthropic states that the process does not compromise the quality, accuracy, or creativity of the output, though the signal may be weaker in factual responses or code due to limited token flexibility.

The rollout is largely driven by the European Union’s AI Act, specifically Article 50, which requires generative AI providers to mark artificial content in a machine-readable format. While the mandate originated in Europe, Anthropic is applying the watermark worldwide. Limitations include the fact that the mark does not prove Claude was the original author—as the text could have been merely processed or summarized by the model—and aggressive paraphrasing may remove the signal.