Anthropic adds invisible watermark to Claude-generated text

Anthropic has rolled out an invisible watermark for Claude-generated text as part of its AI transparency and EU AI Act compliance efforts. The SynthID-Text-based mark is imperceptible to readers, does not affect output quality, and is not a Google ranking signal; a detection API is planned for a private preview.

DEV Community

Open source

Anthropic adds invisible watermark to Claude-generated text

The watermark applies to future Claude models across surfaces including the API and Claude Code
The mark uses a SynthID-Text approach and cannot identify the user or organization
A detection API for authorities is planned as a private preview
Supported files such as images will carry C2PA-based Content Credentials

#Anthropic#Claude

Anthropic adds invisible watermark to Claude-generated text

Anthropic starts embedding token-level watermarks in Claude's text

Anthropic's Production Agent Guardrails: Lint Rules, Fuzzers, and Automated Reviews for Claude-Generated Code

AI Model Watermarking Changes Agent Behavior

Kev: open decision models let AI agents choose without generating text