Anthropic adds invisible watermark to Claude-processed text

Anthropic has introduced an invisible watermark, nicknamed 'Scarlet Letter,' that tags content processed by Claude, including human-written text the model only edited. The mark is not currently visible to end users but flags AI involvement in a given piece of text.

Why it matters: This puts Anthropic ahead of the pack on AI content provenance, a live regulatory and platform-policy issue as governments weigh how to handle AI-generated content. It also raises a scope question: tagging text a human wrote but Claude only edited blurs the line between 'AI-generated' and 'human-authored,' which could complicate how platforms and institutions define AI involvement going forward.