TechAnthropic explains how Claude’s invisible text watermarks will work

Anthropic says Claude’s text-marking system will use a version of Google DeepMind’s open-source SynthID-Text approach. By steering low-stakes word choices with a key, it creates patterns that readers cannot see but authorized detectors can identify. The feature is intended to meet the EU AI Act’s transparency requirements, alongside C2PA support for processed images.

AnthropicClaudetext watermarkingEU AI Act

TThe Verge★★☆☆☆2026-08-17 10:57Original