Claude Users Cancel Subscriptions Over Anthropic's Text Watermark

What Anthropic's New Claude Watermark Does

Anthropic has begun embedding an invisible watermark directly into text produced by some Claude models, a change the AI lab says is intended to identify machine-generated writing without altering meaning, quality or readability. The marker travels with copied text and can survive light editing. Anthropic has applied it globally to output from supported Claude models since August 2 and says it is part of compliance with the European Union's AI Act.

The move has triggered a visible backlash. Dozens of users have posted on X that they canceled subscriptions since Monday, and Business Insider spoke to four users who shared cancellation confirmations. The company said it has not seen an uptick in cancellations since the announcement. Anthropic had 300,000 business customers as of September 2025, and the impact of the watermark on that base is not yet clear.

The complaints centre on authorship rather than quality. Subscribers such as freelance developer Vladislav Rajtmajer, AI consultant Richard Echols and public policy worker Arturo Villarroya said the watermark can remain even when Claude is used only to translate, proofread, fix citations, summarise or lightly edit text the user originally wrote. In coding, translation and client documentation, they worry an invisible AI marker could raise questions about who actually produced the work, or conflict with strict client and academic policies.

Anthropic has said a watermark indicates only that Claude processed text, not that Claude was the original author. The company plans to release a free API that lets users and third parties check text for the marker. Still, some users are already switching to alternatives such as Cursor, Grok 4.6 and other providers, and at least one said the change reinforced his desire to keep his AI layer portable and provider-agnostic.

Why Anthropic's Compliance Watermark Is Hurting Professional Trust

Anthropic's EU Compliance Decision Creates a Professional-Use Friction Point

The watermark is a compliance feature, but its design collides with how many paying users actually deploy Claude. Anthropic says the marker travels with content and may survive edits, including proofreading, translation and summarisation. That means a consultant who writes a document herself and uses Claude only to polish phrasing could still deliver text carrying an AI-processed marker. The practical risk is not that Claude's output reads worse; it is that the marker becomes evidence in environments with strict authorship rules, academic integrity policies or client contracts.

Named Users: Cancellations Driven by Authorship, Not Output Quality

The users Business Insider interviewed describe similar decision logic. Vladislav Rajtmajer, a freelance developer in the Czech Republic, said an AI label on code shipped to clients could raise questions about authorship or trigger contract penalties. Richard Echols, an AI consultant in Georgia, said the mark could appear on documentation when he had written the underlying material himself. Stevan Bogosavljević said having code attributed to Claude would be undesirable for compliance and client policies. These are not quality complaints; they are concerns about control over provenance and professional accountability.

Competitive Fallout: Grok, Cursor and the Push for Provider-Agnostic AI

The backlash is already shaping substitution decisions. Villarroya said he is switching to Cursor and Grok 4.6 for coding and Grok Bot for personal and work organisation. Rajtmajer said the watermark was a signal to evaluate Chinese AI providers and other models. Santiago Alessandro Rivera Martínez, founder of digital agency Directa, said the change reinforced his preference for a portable, provider-agnostic AI layer rather than a workflow built entirely around proprietary software. That points to a broader competitive dynamic: for professional users, provenance control may become a selection criterion alongside model quality.

Steps for Claude Users Managing Client and Compliance Risk

• Check sensitive deliverables before they go to clients. Once Anthropic releases its free detection API, run client-facing code, documents and translations through it to confirm whether a Claude watermark is present, especially if the work was only edited or proofread by Claude.
• Reconsider which workflow steps touch Claude. Users in the story said the mark can survive proofreading, translation and summarisation. If a client contract or academic policy treats AI-processed work as non-conforming, move drafting and editing to non-watermarked tools or document the process explicitly.
• Ask Anthropic about enterprise controls. Business customers and teams on Claude Max should clarify whether their plan has any watermark opt-out or enterprise API setting, because Anthropic has said watermarks apply globally to supported Claude models since August 2.
• Build an explicit AI-use disclosure and provenance record. The cancellations described by Business Insider centre on authorship disputes, not model quality. For each project, record what Claude did — generation, translation, code review or proofreading — so a marker does not become the only evidence in a client or academic review.