Anthropic introduces a new tool and model update that embeds an “imperceptible watermark” into AI-generated text, aiming to help identify content created by its systems. The change is rolled out with new Claude models and is intended to be difficult for people to notice or remove.
The reports describe the watermark as being placed directly in the text produced by the models, rather than added afterward through separate steps. This is presented as a mechanism to reduce the ability to generate AI writing “undetected,” according to the outlets covering the announcement.
While one source frames the update as undermining attempts to use AI novels without detection, the core factual details across coverage focus on the same capability: Anthropic’s new Claude models include an imperceptible watermark embedded into outputs. Both accounts emphasize the technical goal of detection and attribution rather than changes to how users prompt the models.
