**Anthropic** has inserted invisible watermarks into text made by Claude for audiences worldwide, and added signed provenance information to supported files. As the measure is based on an EU regulation joined by about 190 signatories, user backlash is growing amid expanding controversy over copyright and editorial review.
Key content
Machine-readable watermarks that may remain even after copying and pasting are inserted into the text generated by Claude, and the supported files are attached with signed source metadata.
This policy applies worldwide to supported Claude models, while older models are updated sequentially during the transition period.
Critics worry that even if most of the actual text was written by a human, the result could still be marked as AI-generated simply because it received editing or coding support.
Claude watermark introduction
On August 11, Anthropic disclosed how Claude’s AI-generated outputs would be marked, saying it is part of implementing the transparency framework under the European Union’s AI Act (AI Act). The Claude models released in the EU since August 2 have the watermarking feature built in from the beginning, and the company is also expanding the feature to models previously released.
This system works by directly embedding watermarks into the generated text at a level that users cannot perceive. Image files in supported formats such as SVG, PNG, and JPG are given source metadata with digital signatures according to the C2PA standard. Anthropic emphasized that this text watermark “does not change meaning, quality, or readability.” According to reporting by Forbes, users cannot turn the feature off selectively; it is enabled by default.
However, there are technical limits to watermarking. With extensive rewriting, heavy editing and paraphrasing, translation, or mixing the text with other writing, signals within the text can be lost. File source metadata may also disappear during format conversion, re-saving, or screenshot (capture) processes.
Reference article: XRP Breaks Below $1 While Whales Quietly Scoop Up 380M Tokens
User backlash spreads
Immediately after the announcement, criticism spread quickly on platforms such as **X (formerly Twitter)** and Reddit. In particular, users who use Claude as an editing and sentence-polishing tool and developers who use it for writing and reviewing code voiced complaints—“Are results that people wrote themselves going to be marked as AI-generated?”
Radio host and blogger **Erick Erickson** criticized the idea that watermarking would be applied just because the text he personally wrote had only been through Claude’s editing. He called it “ridiculous.”
This concern highlights the key difference in the guidelines Anthropic itself has presented. The fact that a watermark was detected is only a signal that the content “may have been processed by Claude,” not that “Claude is the original author.” The company also states that it can produce outputs with watermarks even if the work involves editing, translation, summarization, or file conversion based on existing human-created content.
Meanwhile, the EU’s related “code of practice” itself is closer to a voluntary norm, but the transparency obligation set out in Article 50 of the AI Act became legally effective starting August 2. As of the end of July, about 190 organizations, including Google, Meta, Microsoft, OpenAI, and Anthropic, had signed the code.
A “source attribution” trend spreading across the industry
The change is not limited to Claude. OpenAI has documented that for currently supported images, it is using C2PA metadata and the SynthID watermark in parallel, and that it applies SynthID to supported audio. However, in the currently available source-verification tools, text is not included in the formats being detected.
As big tech and major AI companies begin rolling out watermark and source-metadata systems in the wake of EU regulations and voluntary codes, debates over copyright, responsibility, and where to draw the line on what counts as “AI-generated” outcomes in human–AI collaboration are expected to heat up further.
Next article: Apple Pay Chief Jennifer Bailey Retires After 25 Years At The Company