Claude now stamps some of its answers with invisible ink. Anthropic announced this week that certain Claude versions will add a hidden watermark to the text they generate.
Anthropic insists the stamp leaves a response's substance, caliber, and legibility untouched. It travels with copied text and can survive some edits.
The company notes that a detected mark means Claude touched the text, not that Claude invented it.
For some paying users, that's a dealbreaker.
Arturo Villarroya told Business Insider: "I do use it for everything, mostly coding apps and tools." He also leans on Claude for research, sourcing, and corrections at work.
He canceled his Claude Max plan Wednesday. He worried the mark could stick even when Claude only translated text, cleaned up grammar, or gave his original writing a light summary.
In academic or professional settings with strict AI-use rules, that label could become a problem, he said. For his coding needs, he has moved to Cursor and a separate service for his professional assignments.
Vladislav Rajtmajer, a Czech Republic-based freelance developer, canceled Tuesday. His AI work covers code reviews and translations, and he worried that an AI label on client code could raise authorship questions or trigger contract penalties.
He called it a "signal to look at Chinese AI providers and learn to work with other models."
Richard Echols, an AI consultant in Georgia, canceled Wednesday. He uses Claude for business documentation and worries the mark appears even when he wrote the material himself.
"Even if you edit your own work that you wrote, they add the watermark anyway," he said.
Stevan Bogosavljević, a senior software engineer based in Montenegro, had already left in July, citing worsening service quality, restricted access to leading models, and better options elsewhere. He said the watermark confirmed his decision.
He said an invisible watermark on shipped code "definitely would be not desirable."
Santiago Alessandro Rivera Martínez, founder of digital agency Directa, canceled this week too. He said building a workflow around proprietary software means the provider can change the rules later.
"I would rather have the AI layer of my work be portable and provider-agnostic," he said.
Not everyone thinks the watermark is bad. Aadit Sheth, cofounder of The Narrative Company, posted on X that "audiences should know if words reflect a person's own thinking, and watermarks help ensure that."
Watermarks could also keep AI systems from training on AI-generated content, which can cause them to degrade. Anthropic added the watermark to comply with the European Union's AI Act.
The company also plans to release a free API in September 2025, a tool that lets anyone check text for Claude's mark. Other big players are already labeling AI content.
Google deploys a watermarking tool called SynthID, while OpenAI applies SynthID to supported images and audio. X adds a "Made with AI" tag to content it thinks is AI-generated or manipulated.
These industry-wide efforts aim to prevent misuse and maintain transparency, but they also raise concerns about unintentional labeling of human work.
If you use Claude, the watermark may already be part of your text. You can't see it, and it won't change what you wrote, but soon anyone can check for it with a free tool.
The bigger question isn't how the technology works. It's whether you're comfortable with a hidden mark traveling with your writing, even the parts you wrote yourself.
The people who left answered no. Claude Max costs $100 a month, and they didn't want a label they can't see riding along.
But these were power users paying top dollar.
When that kind of user starts walking, you don't need 300,000 of them for the message to be heard.
