Anthropic has begun introducing invisible, machine-readable watermarks into text produced by supported Claude models, prompting a heated debate among users over AI transparency and ownership of AI-assisted work.
The change follows new transparency requirements under Article 50 of the EU AI Act, which took effect on August 2, 2026. AI providers are required to make generated or manipulated content detectable using machine-readable markings. Anthropic has also signed the EU’s voluntary Code of Practice designed to help companies demonstrate compliance.
How Claude’s Watermark Works
Anthropic says Claude models launched on or after August 2 support marking from launch, while the company is also working to introduce the technology to older models.
The watermark is woven directly into generated text and is invisible to users. It does not alter the meaning or readability of the response and can remain attached when text is copied and pasted or lightly edited. The system applies worldwide across supported Claude products, including Claude, Claude Code, Claude Cowork and the API.
Some Claude Users Are Angry
The move has drawn criticism on Reddit. One user, whose account was only around three weeks old, argued that ordinary users could be unfairly exposed for using Claude to reorganize writing, summarize material or overcome writer’s block, while more technically skilled users could potentially remove evidence through rewriting or paraphrasing.
Other Reddit users rejected that argument, pointing out that submitting Claude-generated work as entirely human-created could itself be misleading.
Another critic described the watermarking system as “unethical” and argued that users often provide the ideas, instructions, context and repeated refinements, with Claude acting only as a tool. Other commenters responded that the watermark is intended to identify AI involvement rather than claim ownership of the finished work.
Debate Over AI Transparency
Some criticism has focused on the perceived irony of AI companies marking generated material despite training frontier models using large amounts of existing human-created content.
Others have strongly supported the policy, arguing that people should have a way to determine when material has been generated or processed by AI.
Anthropic acknowledges that its system is not definitive. A watermark can appear on material Claude only proofread, translated, or summarized, while heavy editing, paraphrasing, translation, or combining the text with other writing can make a watermark undetectable. A detected mark therefore indicates that Claude may have processed the content rather than proving Claude originally authored it.
The post Users Angry as Claude Outputs Can Now be Caught by AI Tools appeared first on ProPakistani.
