Anthropic embeds invisible watermark in Claude text under EU AI Act

Anthropic has started embedding an imperceptible, machine-readable watermark in Claude-generated text as part of its EU AI Act compliance. The marker preserves meaning, adds no tokens and survives copying, but is not a Google ranking signal.

• Watermark operates at model level across API, Claude, Claude Code and cloud partners
• Marker is invisible to readers, adds no tokens and persists when text is copied
• New Claude models in the EU mark AI content from day one since August 2, 2026
• Anthropic runs a detection API in private preview for authorized entities