Anthropic Embeds Imperceptible Watermarks Into Claude Text Output

Anthropic is embedding an "imperceptible watermark" into Claude-generated text at the model level — woven into output without changing meaning, quality, or readability. The watermark survives copy-paste and some editing, unlike file-level metadata. Anthropic is working to enable third-party detection, with technical documentation forthcoming. New Claude models will ship with markings from day one; existing models are a work in progress.