Anthropic Deploys Invisible Watermarks on Claude to Meet EU AI Transparency Rules

Key Points

• Anthropic will embed an invisible watermark in all text generated by Claude to comply with EU AI transparency rules.
• The watermark works by using a secret key to influence word selection, leaving a hidden pattern detectable only with the key.
• Output quality, speed, and token cost remain unchanged; the watermark does not add visible artifacts.
• Anthropic will release an API that can decode the watermark and confirm Claude's involvement in a text block.
• Limitations include inability to differentiate between original generation and light editing, and reduced effectiveness on very short text.
• Code receives less watermarking because exact output leaves little room for word‑choice variation.
• Claude‑generated images will carry a cryptographically signed metadata note indicating AI origin.
• Watermarks apply to all Claude models released after Aug. 2, with older models slated for updates in the coming months.