Anthropic recently decided to watermark Claude’s output. Insert an invisible code into your chatbot’s edit text to indicate it was generated by an AI.
Anthropic introduced this new policy to meet the transparency provisions of the EU AI Act. The provision now requires technology companies to label content generated or edited by AI in a way that computer systems can identify it. But while European regulators may be happy, some AI users clearly aren’t.
You don’t have to look far on Reddit to find evidence of growing frustration, but other posters on the site disagree. One of the more historical posts I found was from a user named visionode, especially since the account has only been established for three weeks. According to visionode, the new watermarking system is a strict conspiracy aimed at victimizing innocent chatbot users around the world.
Visionode’s basic argument seems to be that while savvy Claude users may be able to hide evidence of AI use by rephrasing or cleaning the output through other AI services, the average Claude user will get caught.
“Who will get caught? You. The student who used Claude to rearrange a paragraph. The journalist who asked an AI to summarize a 200-page transcript. The writer who had a creative block and asked for a synonym. They come out of the process with a digital tattoo on their forehead.”
I don’t mean to diminish visionode’s ire, but these aren’t the best examples. A journalist who asks an AI to summarize a 200-page transcript won’t care about the watermark on the summary unless he or she is copying and pasting it verbatim into an article. This is clearly unethical and should not happen.
Equally unethical is a student who asks him to “reorganize a paragraph” and then copies and pastes Claude’s output into his essay.
Other Redditors didn’t particularly support the poster’s outrage.
One poster simply commented: “Eat this guy up a lot.”
Another asked OP to “take a deep breath.”
Visionode wasn’t the only one dissatisfied. Other disgruntled campers called the watermarks “unethical” and “offensive” and claimed they had done most of their work by using Claude. For them, chatbots were just a “tool” that made their hard work easier.
“I provided the instructions, context, decisions, and countless improvements. That tool was Claude. If Claude starts watermarking the code and what it produces, what credit can it claim?” the poster asked.
Once again, other users latched onto the critic.
One user countered, “That’s not claiming credit.” “Being able to detect AI-generated output is important because it can pose a risk in a variety of situations.”
“My brother couldn’t even complain about Claude without writing using Claude,” another joked.
Some critics are making a more nuanced argument for Anthropic’s new policy, eschewing the narrative of victimhood.
For example, one poster complained about the common hypocrisy of fabricating someone else’s work and watermarking the generated editorial work. “I think this is a very sinister direction,” one user said. “I’m not using Claude to write anything, but having an AI watermark your work is horrifyingly ironic considering how many frontier models come from training data.”
But in general, users tend to support watermarking systems as a sensible way to track material generated by algorithms. A user in another thread said, “There’s literally no good argument why this isn’t a good idea.” “The only reason you don’t want this is to lie to people.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
