Anthropic has begun embedding invisible watermarks in text generated by its Claude AI models to comply with the European Union’s AI Act Transparency Code effective August 2, 2026 [1, 2]. The watermarking is applied at the model level, meaning all new Claude models released after that date include the watermark, with older models to be updated over the coming months [1, 2]. Anthropic stated the mark does not alter the meaning, quality, or readability of AI-generated text [3, 2].
The watermark uses subtle statistical patterns in word choices detectable only with a secret key, based on a 2024 Google DeepMind research paper [2]. Embedded imperceptibly in the text itself, the watermark travels with copied or pasted content and may persist after some editing [1, 3, 2]. According to Anthropic’s support page, "Because the watermark is part of the text, it will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing. Watermarking will be applied at the model level, which means it will be present no matter which Claude product or surface the text comes from" [1].
The watermark applies globally as Anthropic lacks the technology to restrict it by region [2]. The move follows the EU AI Act’s Transparency Code, which took effect August 2 and mandates AI companies disclose machine-generated content [1]. Anthropic confirmed watermarking implementation on August 11, followed by user discontent the next day as complaints surfaced about privacy and usability issues [1, 4].
Some Claude users worry the watermark could expose AI assistance in professional or academic work, creating risks and stigmas. One anonymous Reddit user wrote, "Who will get caught? You. The student who used Claude to reorganize a paragraph. The journalist who asked the AI to summarize a two-hundred-page transcript. The writer who had creative block and asked for synonyms. Those guys come out of the process with a digital tattoo on their forehead" [4]. Former Claude Max subscriber Arturo Villarroya expressed concern that watermarking "could remain even when the underlying writing was original and Claude was used only to translate, check spelling, fix academic citations, or lightly summarize a passage" [3].
From August 13 to 14, multiple users publicly canceled Claude subscriptions citing discomfort with watermarking [3]. Despite this, Anthropic reported no significant jump in cancellations and maintains they have about 300,000 business customers as of September 2025, with Claude Max subscriptions starting at $100 per month [3].
Anthropic defended the watermark on August 14 in a blog post, emphasizing its necessity for EU regulatory compliance and insisting the feature does not compromise output quality or meaning [2]. Supporters argue this is a needed transparency tool, while critics see it as punitive or invasive [4, 2].
Anthropic plans to roll out watermarking to older Claude models over the next few months, continuing enforcement of the EU AI Act requirements globally [1, 2].
