Anthropic's Claude watermark prompts subscription cancellations

Anthropic announced that certain Claude models will embed an "imperceptible watermark" in generated text, a marker that persists when the content is copied, pasted, or lightly edited. The company says the watermark does not affect meaning, quality, or readability and is intended to help identify AI-processed material.

Since the rollout on August 2, a number of subscribers have reported canceling their Claude Max plans, which start at $100 per month. Arturo Villarroya, a public-policy professional, said he stopped using the service because he felt uneasy about the watermark potentially appearing in work such as coding, research, and document editing. He is switching to alternative tools for both professional and personal tasks.

Freelance developer Vladislav Rajtmajer expressed similar worries, noting that an AI label on code delivered to clients could raise questions about authorship or breach contract terms. He described the watermark as a signal to consider other AI providers.

AI consultant Richard Echols added that the watermark could appear even when he edited his own material, undermining confidence in the output. Senior software engineer Stevan Bogosavljević, who had already cited declining service quality, said the watermark reinforced his decision to leave, fearing it could affect compliance and client policies.

Anthropic clarified that the watermark indicates Claude processed the text, not that it authored the original content, and that the mark can remain after proofreading, translation, or summarization. The company said the feature supports compliance with the European Union’s AI Act and that an API will allow users to detect the watermark themselves.

Industry observers note that watermarks aim to increase transparency and may help prevent AI-generated content from being inadvertently used to train future models. Similar technologies are employed by Google and OpenAI, and X adds a "Made with AI" tag to identified posts.