Anthropic to Watermark All Claude AI Output Starting This Week
The company will embed invisible signatures in text and images to comply with EU transparency rules, sparking immediate user backlash.
Anthropic announced Tuesday it will begin embedding invisible watermarks in all content generated by its Claude AI assistant, a move that immediately triggered widespread criticism from users concerned about their work being flagged as AI-generated.
The new policy applies globally to all Claude models and products, including Claude Code and Claude Cowork, according to details first reported by Forbes. Users cannot opt out of the watermarking system.
How the watermarks work
Claude-generated text will carry an invisible watermark that persists when content is copied and pasted to other locations. Images and other files created by the AI will include digitally signed metadata.
Anthropic acknowledged the system has limitations. Text watermarks may disappear if writing undergoes heavy editing, and file metadata can be stripped through format changes, screenshots, or other alterations. The company said it will publish additional information about how to detect the watermarks in practice.
Why it matters
The watermarking requirement stems from the EU's Code of Practice on Transparency of AI-Generated Content, which mandates marking and detection systems for AI-created media. Nearly 200 companies have signed the code, including Meta, Microsoft, and OpenAI. Google has watermarked AI-generated images since 2023 and has expanded the practice to text, audio, and video. OpenAI also watermarks images but has declined to watermark text output from ChatGPT despite having the technology available for years, citing concerns about false positives and competitive disadvantage.
For enterprise users, the policy creates new compliance considerations around content provenance and raises questions about how watermarks might affect code repositories, documentation systems, and collaborative workflows.
User backlash intensifies
Criticism erupted quickly on social media platforms, with users calling the policy "hugely problematic" and declaring an end to "the era of unprovable AI writing."
Many objections came from people who say they use Claude solely for proofreading their own writing. Radio host and blogger Erick Erickson posted on X that he had switched from Grammarly to Claude for proofreading, but now worries his original writing will be marked as AI-generated.
Software developers expressed concern that cryptographic signatures embedded in code could degrade output quality or create technical complications.
The announcement also sparked debate about attribution. Some users argued that humans who provide instructions and context should receive credit for AI-generated work, while others contended that watermarks simply acknowledge the role of AI in content creation without assigning credit.
Regulatory context
The policy change brings Anthropic into compliance with European regulations as AI transparency requirements expand globally. The EU code provides specific rules for marking AI-generated content and labeling deepfakes and manipulated media.
OpenAI has reportedly debated deploying text watermarking internally for years but has resisted due to concerns about accuracy, circumvention methods, and the risk of losing users to competitors, according to the Wall Street Journal.
Details were first reported by Forbes staff writer Mary Roeloffs.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
