Some Claude users criticize Anthropic's new AI-output watermarks

Anthropic has started inserting invisible code into Claude's editorial text outputs to identify them as AI-generated. The change was made to comply with the European Union's AI Act Transparency Code, which requires technology companies to label content that has been AI-generated or edited in a way identifiable to computer systems.

Mixed reactions online

While European regulators may welcome the move, reactions among Claude users have been divided. On Reddit, several posts have criticized the new policy. One user, whose account was only three weeks old, argued the watermarking system unfairly targets ordinary users, citing examples such as students asking Claude to reorganize a paragraph or journalists asking it to summarize a lengthy transcript. Other Reddit users pushed back, pointing out that watermarks only become a problem if AI-generated text is copied directly into final work without disclosure — something already considered unethical.

Another critic called the watermarks "unethical" and "disgusting," arguing that Claude was merely a tool while the user supplied the instructions, context, and refinements that constituted the bulk of the work. Other commenters responded that the watermarks aren't about claiming credit, but about enabling detection of AI-generated content in situations where it could pose risks.

A question of irony

Some critics raised a more nuanced point, noting an apparent contradiction: a company whose models were trained on large amounts of other people's work is now marking its own output as identifiable. Still, overall sentiment among users leaned toward supporting the watermarking system as a reasonable way to track algorithmically generated material, with some arguing the only reason to oppose it would be a desire to mislead others.