Claude AI watermark concerns push some paid users to cancel subscriptions
The new Claude AI watermark is facing resistance from some paying users who worry that an invisible marker could remain attached to professional work even when Claude only helped edit it.
Anthropic has introduced an imperceptible watermark into text from supported Claude models. The marker can travel with copied text and may survive some editing, according to the company.
Business Insider spoke with four people who said they had canceled Claude subscriptions. Other users posted similar claims online.
Anthropic, however, told the publication it had not observed a broader increase in cancellations since announcing the feature. That distinction matters. A visible social media backlash does not automatically show that a significant share of Claude’s customer base is leaving.
The Claude AI watermark is designed to identify text that has passed through Claude. That is different from proving Claude originally wrote the material.
According to Anthropic’s official explanation of Claude’s text watermark, the watermark is intended to indicate the likelihood that Claude was involved in producing or processing text. Anthropic says the system is being introduced as part of compliance with the European Union’s AI Act.
A developer could write code independently and use Claude for a review. A writer could submit original text for grammar corrections. In both cases, a detector might indicate that Claude processed the material without explaining how much of the final work came from the AI.
Business Insider reported that Claude Max starts at $100 per month. One user interviewed by the publication said the watermark made him uncomfortable because he uses Claude for coding, research and corrections.
There is a policy reason behind Anthropic’s move. The company says the watermark helps it comply with the European Union’s AI Act as transparency rules for generated content develop.
The issue also fits a wider shift toward autonomous Claude systems. The AI Decode recently covered a Claude AI agent that hacked a gym booking system, showing how Claude is increasingly being used for tasks that go beyond ordinary chatbot conversations.
Watermarking is not unique to Anthropic. Other technology companies have introduced methods for identifying AI-generated images, audio and other media. But reliable text watermarking has been particularly difficult because text can be edited, translated or rewritten.
Anthropic’s wider challenge is maintaining trust as Claude expands. The AI Decode has also examined how Claude Code is powering recurring AI loops, another example of Claude moving deeper into professional software workflows.
The Claude AI watermark debate therefore comes down to a difficult distinction: transparency about AI involvement versus an assumption of AI authorship. What matters next is whether detection systems can communicate that difference clearly enough for employers, schools and clients to use them responsibly.
