Anthropic sparks user backlash with Claude text watermarks

Anthropic recently implemented a new policy to inject invisible watermarks into the text generated by its Claude chatbot. This cryptographic signature is designed to identify the editorial outputs as machine-generated. The company introduced the feature to comply with the European Union AI Act's Transparency Code, which mandates that tech companies clearly label content that has been generated or altered by artificial intelligence systems.

While the move satisfies European regulators, it has provoked a wave of anger among some Claude users. On platforms like Reddit, critics have voiced concerns that the hidden tracking code will expose their reliance on the chatbot in professional and academic settings. One disgruntled user described the watermarking system as a "digital tattoo" that would unfairly penalize average users who rely on Claude for basic tasks like reorganizing paragraphs or summarizing long transcripts.

Other critics labeled the watermarks "unethical" and "disgusting," arguing that because they provide the prompts, context, and refinements, they should hold the credit for the final output. Some users also pointed out the irony of a frontier AI model watermarking its outputs when the model itself was trained on vast amounts of web data created by humans.

However, many in the AI community have defended Anthropic's decision. Supporters argue that watermarking is a necessary safety measure to prevent deception and track algorithmic content. Defenders of the policy noted that the only reason to oppose such transparency is an intention to mislead employers, teachers, or readers about the true origin of the work.