𝐀𝐧𝐭𝐡𝐫𝐨𝐩𝐢𝐜 𝐑𝐞𝐯𝐞𝐚𝐥𝐬 𝐇𝐨𝐰 𝐂𝐥𝐚𝐮𝐝𝐞’𝐬 𝐍𝐞𝐰 𝐀𝐈 𝐖𝐚𝐭𝐞𝐫𝐦𝐚𝐫𝐤𝐬 𝐖𝐢𝐥𝐥 𝐖𝐨𝐫𝐤

Anthropic has shared more details about its new watermarking system for AI-generated text in Claude, designed to help identify content created by AI while keeping the watermark invisible to readers.

The system will use SynthID-Text, developed by Google DeepMind, and Anthropic plans to offer a watermark detection API. Light editing may not remove the watermark, while a complete rewrite likely will.

Anthropic says the technology will have little impact on code, although watermarks could appear in areas such as code comments where wording choices are more flexible.

The move comes as AI companies respond to the EU AI Act’s transparency requirements, signaling a broader shift toward making AI-generated content more identifiable.