Anthropic to use steganography for Claude text watermarking
August 16, 2026
Anthropic will implement watermarking by using steganography to influence token selection during inference. This method creates detectable linguistic fingerprints in generated text to comply with EU regulations.
HOW THIS AFFECTS YOU
●
builderYou may need to account for altered token distributions in downstream text processing.
●
policyThis implementation seeks to meet regulatory requirements for AI content identification.