Prime Highlights-
- Anthropic embeds an imperceptible watermark into Claude-generated text to support EU AI Act transparency requirements.
- Publishers and schools gain a new tool to investigate AI-generated writing amid recent authorship controversies.
Key Facts-
- Claude models launched from August 2 onward carry the watermark by default, with older models to follow.
- Heavy editing, paraphrasing, translating or mixing Claude’s output with other text can make the watermark undetectable.
Background-
Anthropic announced that its newer Claude models will embed an imperceptible watermark directly into AI-generated text, a move aimed at making it easier to identify writing produced by the company’s chatbot.
The watermark does not alter a text’s meaning or readability, travels with it when copied and pasted, and can persist through some editing, according to the company.
A watermark refers to a faint pattern or marker embedded in digital content to help establish ownership and deter unauthorized copying. Anthropic said the rollout supports its transparency commitments under the European Union AI Act. Claude models launched from August 2 onward will carry the marking by default, with the company working to extend the feature to older models.
The watermark applies to Claude-generated content globally, including access through cloud providers, and Anthropic plans to give third parties tools to detect it.
The development lands as the publishing industry works through concerns around AI-generated manuscripts. A literary agent previously withdrew support for a crime novel after questions arose over possible AI involvement, despite the book drawing bids from fourteen publishers before the deal closed.
The author denied using AI. In a separate case, a publisher pulled a horror novel following AI-generated writing allegations, with the author saying a freelance editor had introduced AI material without her knowledge.
Anthropic’s new watermark could give publishers, schools and universities an additional tool to investigate such disputes. The company acknowledged limitations, noting that heavy editing, paraphrasing, translating or blending Claude’s output with other writing can make the watermark undetectable, and that detecting a watermark does not confirm Claude originally authored a piece, since even proofreading or translation can leave a trace.
Anthropic becomes the second major AI lab to introduce text watermarking, following Google DeepMind, which began watermarking text and video generated through its Gemini products using SynthID technology, after first releasing an image version of the tool.
