Anthropic has recently decided to add watermarks to the generated content of Claude, which means embedding invisible codes in the text produced by the chatbot to indicate that the content was generated by AI. This policy is aimed at meeting the transparency requirements of the EU AI Act, which mandates tech companies to mark AI-generated or edited content with computer-readable identifiers. While regulators may be satisfied, some AI users are clearly not happy, and discussions have already heated up on Reddit.

One of the most heated posts came from user visionode, whose account has been active for less than three weeks. He refers to this watermark system as a "strict conspiracy" designed to make innocent chatbot users victims. His core argument is that experts who are familiar with Claude can rewrite or use other AI services to hide their traces, while ordinary users are left vulnerable.

The "Digital Tattoo" Debate: Who Are the Real Victims?

visionode paints a picture of "victims": "Who will get caught? You - the student who asked Claude to reorganize text; the journalist who used AI to summarize 200 pages of interviews; the writer who asked AI for synonyms. These people will all end up with a digital tattoo on their forehead." However, these examples were quickly refuted: if a journalist only uses AI to summarize material, the watermark would not have a real impact unless he copies the AI summary directly into his article; a student who copies AI-revised paragraphs into a paper already violates academic ethics. Other users did not support his anger, with some leaving a comment like "look at this guy" and others advising him to "take a deep breath."

Another critic called the watermark "unethical" and "disgusting," emphasizing that he provided instructions, background information, and numerous revisions during his use of Claude, asking, "Claude is just a tool, what credit is it taking?" However, opponents responded that the watermark is not about taking credit but about traceability in scenarios where AI-generated content poses risks. Some users also raised deeper concerns: since cutting-edge AI models are trained on others' works, adding watermarks to AI-generated content carries a "frighteningly ironic" meaning. Nonetheless, overall, most users still support the watermark system, with one user's summary being particularly representative: "The only reason not to want such a mechanism is if you want to lie to others."