Anthropic's Claude watermarks spark backlash from users worried about detection

Anthropic recently implemented a new system to watermark Claude's outputs by inserting invisible code into the chatbot's text. This code marks content as AI-generated in a way that computer systems can detect, even though the watermarks remain imperceptible to human readers.

The decision stems from Anthropic's need to comply with the EU AI Act's Transparency Code, which requires tech companies to label AI-generated or AI-edited content in a manner identifiable to automated systems. This regulatory requirement pushed Anthropic to adopt a technical solution rather than relying solely on visible labels or terms of service.

The policy has generated mixed reactions, particularly on Reddit. One user named visionode, whose account is three weeks old, argued that the watermarking system would disproportionately catch average users while savvier ones could hide their AI usage by paraphrasing or running outputs through other AI services. Visionode framed the watermarks as a "draconian conspiracy" and gave examples of what they saw as innocent uses: "The student who used Claude to reorganize a paragraph. The journalist who asked the AI to summarize a two-hundred-page transcript. The writer who had creative block and asked for synonyms." However, other Redditors pushed back, noting that a journalist summarizing a transcript for reference would not be harmed by a watermark unless they copied the summary verbatim into a published article—which would be unethical. Similarly, a student who copies and pastes Claude's reorganized paragraph into an essay without attribution is committing academic dishonesty, not simply using a tool.

Another critic called the watermarks "unethical" and "disgusting," arguing that because they had provided instructions and refinements, Claude was merely a "tool" and should not be watermarking their work. A commenter responded by clarifying that the watermarks are not about Claude claiming credit but rather enabling detection of AI-generated outputs to manage the risks such content can pose in various situations. One user also pointed out the irony of complaining about Claude without noting they were using Claude to write their complaint.

A more substantive criticism raised the question of hypocrisy: watermarking an editorial product that was itself generated by training on others' work without explicit consent. However, most Reddit users supported the watermarking system, with one commenter stating, "There is literally no good argument for why this isn't a good idea. The only reason you wouldn't want this is to lie to people."