Anthropic Adds Invisible Watermark to Claude to Tackle AI Content

Anthropic says new Claude models will embed an invisible watermark in AI-generated text to boost transparency and fight misuse. The watermark won't change meaning or readability, and it survives copy-paste or light edits. Models released after August 2, 2025, support it by default, with older models getting updates. It applies to all Claude content worldwide, including via cloud services. Anthropic plans to give third parties tools to spot these watermarks.

This could shake up publishing. Recently, books like "Call Me, I'll Hide the Body" and "Shy Girl" were pulled over AI-use suspicions. The watermark gives publishers and schools a new way to check, but it's not foolproof—heavy rewriting, translating, or mixing text can hide it. Anthropic is the second big AI lab to do this, after Google's SynthID for Gemini.