Anthropic to embed invisible watermarks in Claude-generated text using Google’s SynthID-Text

Anthropic announced it will embed invisible watermarks in text generated by its Claude models using a version of Google DeepMind’s open-source SynthID-Text watermarking system. The company described the approach as creating detectable patterns in wording probabilities without altering the meaning or quality of the output. Anthropic framed the feature as part of its compliance strategy for the European Union’s AI Act, which mandates machine-readable marks for synthetic text, images, audio, and video to indicate artificial generation or manipulation.

Anthropic emphasized that the watermarking will not increase costs for users or degrade the quality of Claude’s responses. The system works by influencing low-stakes word choices during generation—moments where multiple plausible continuations exist—using a seeded randomness derived from a key. This leaves an undetectable pattern in the text that can be verified by anyone with the corresponding key, while remaining imperceptible to human readers.

The EU AI Act’s transparency requirements also apply to other major developers, and Anthropic noted that Google’s Gemini chatbot has supported the SynthID-Text solution since 2024. OpenAI has not publicly detailed text watermarking plans for ChatGPT in its AI Act compliance roadmap, but will be subject to the same legal obligations.