Anthropic has started adding invisible watermarks to text produced by its Claude chatbot. The code is embedded directly into the output and marks it as AI generated, even though it cannot be seen by the human eye.
The change comes as a response to new rules in Europe. The EU AI Act's Transparency Code now requires tech companies to label content that has been created or edited by AI in a way that computer systems can identify.
European regulators appear satisfied with the update. Many Claude users are not.
Why Anthropic Added Watermarks
The watermarking system flags any text that comes out of Claude, whether it is a full essay or a short paragraph. This applies even when a person only asks the chatbot to tweak a sentence or find a synonym.
Anthropic built the feature to meet legal requirements set by European regulators. The rule is meant to help people tell the difference between human written and AI written content.
Reaction to the policy has been mixed on sites like Reddit. One user, posting under the name visionode, argued the system unfairly targets people who use Claude for small, everyday tasks.
Visionode's post claimed that students, writers, and journalists using the tool for minor edits would end up with what they called a digital mark against them. The post suggested that only advanced users who know how to disguise AI text would escape detection.
That argument did not sit well with other users on the platform. Several pointed out that using AI text without disclosing it is already considered unethical, regardless of watermarks.
One user responded by telling the original poster to take a breath. Another commented that the reaction was overblown.
A separate critic took a different angle in a Reddit post. This person argued that they had done most of the work themselves, using Claude only as a supporting tool during the writing process.
They asked what exactly the AI was claiming credit for, if a human provided the direction and made the final decisions. This post also drew pushback from other commenters.
Mixed Reactions From Users
One reply explained that the watermark is not about claiming credit at all. Instead, it exists to help identify AI generated text because of the risks that unlabeled AI content can create.
Another commenter pointed out an irony in the original complaint, since the post itself appeared to have been written with help from Claude.
Not every criticism focused on personal inconvenience. One user raised a separate concern about fairness, noting that AI models are trained using large amounts of existing human written material.
This person argued it seemed ironic for a company to watermark its own AI output, given how frontier AI models are trained on other people's work in the first place.
Despite these criticisms, most commenters appeared to support the new policy overall. One user wrote that there was no strong argument against the feature, adding that the only reason to oppose it would be to hide AI use from others.
The debate remains active on Reddit as of this week, with new comments continuing to appear on multiple threads discussing the policy.
