Anthropic and Google Have Quietly Embedded Invisible Watermarks in AI Text

nikvassev · x · 2026-08-12

The irony of AI text watermarking is hotly debated: people claim they don't use AI for writing, yet are highly upset about watermarking.

According to the quoted article, starting August 2, Anthropic's new Claude models will weave an invisible watermark into generated text and publish the detection method. Furthermore, Google has actually been embedding hidden signatures in Gemini's outputs since 2024.

Related event: Claude Introduces Invisible Watermarks and Metadata to Comply with EU AI Act(83 posts)→

Original post →

More from Models

Models channel →