Anthropic recently decided to watermark Claude’s outputs, inserting invisible code to satisfy the EU AI Act’s Transparency Code. While regulators may be happy, some Reddit users are not. User visionode claimed the system is a draconian conspiracy to catch average users, like students or journalists who ask Claude to reorganize paragraphs or summarize transcripts. However, other posters argued that copying the output verbatim into an article or essay is plainly unethical and shouldn’t happen.

Another critic called the watermark “unethical,” arguing that they provided the instructions and context, so Claude was just a tool. They felt the watermark claimed credit for their work. But others countered that watermarking is about detecting AI-generated outputs due to risks, not claiming credit. Some pointed out the irony of watermarking content trained on others’ work. Overall, most users support the policy, saying there is no good argument against it unless you want to lie.

Anthropic’s move fulfills the EU AI Act’s requirements, yet the backlash highlights tensions around AI transparency and user intent. The article notes that Lucas Ropek, a senior writer at TechCrunch, covers AI.