What happened?

Anthropic recently made the decision to watermark Claude's outputs — inserting invisible code into the chatbot's editorial text that marks it as AI-generated.

The company rolled out this policy to satisfy the EU AI Act's Transparency Code, which now requires tech companies to label content that has been AI-generated or edited in a manner identifiable to computer systems. European regulators may be happy about that; some AI users are decidedly not.

Where the backlash is

One need look no further than Reddit to find evidence of brewing discontent — though other posters on the site are not in agreement. One of the more histrionic posts came from a user named visionode, whose account is notably only three weeks old. According to visionode, the new watermarking system is a draconian conspiracy designed to victimise innocent chatbot users worldwide.

The basic argument is that while savvy Claude users may be able to hide evidence of their AI usage by paraphrasing or otherwise cleaning their outputs through other AI services, the average user of Claude will be caught.

The examples given

The examples listed in the post were:

  • The student who used Claude to reorganise a paragraph
  • The journalist who asked the AI to summarise a two-hundred-page transcript
  • The writer who had creative block and asked for synonyms

TechCrunch's assessment is that these are not the best examples: a journalist asking AI to summarise a two-hundred-page transcript is not going to be bothered by a watermark attached to that summary, unless they are copying and pasting the summary verbatim into their article — which is plainly unethical and should not be happening. The same holds for a student who pastes Claude's output into an essay after asking it to "reorganise a paragraph".

What other users said

Other Redditors were also not particularly supportive of the poster's outrage. One commented simply, "Get a load of this guy"; another asked the original poster to take "a deep breath". Visionode was not the only one complaining, but the argument did not run in one direction.

The scope of the rule

The EU AI Act's transparency clause does not ask for content to be labelled for the human eye but for it to be marked in a way identifiable to computer systems. That distinction explains why the watermark is invisible: the aim is not to show the reader a notice but to let a tool test the text afterwards.

That the obligation sits with the model provider matters too. The party applying the rule is not the person using the text but the company producing it, so the user has no choice over this marking.

Why does it matter?

The backlash itself says something about what watermarking technology does. A transparency tool has its most visible effect on those who want to hide — and that exposes the distance between the aim of the regulation and the expectation of the user.

The justification for the EU rule is that a reader should be able to know how a text was produced. The justification for the complaint is the opposite: that how the text was produced should not be known. Both sides are talking about the same consequence of the same feature.

What is not settled

There is no measurement of how widespread the backlash is; the report rests on individual posts, and the account behind the most prominent one is three weeks old. Under what conditions the watermark can actually be detected, and how robust it is, also falls outside the scope of this report.