Top News

Anthropic Claude Watermark: Following the emergence of AI tools like ChatGPT and Claude, a wide variety of AI-generated content has appeared online. This has made it difficult to distinguish whether a text or file was created by a human or by AI. To address this issue and ensure transparency, the AI ​​company has updated its support page, announcing that it is adding machine-readable marks to Claude's output to comply with the transparency rules outlined in the European Union's AI Act. Essentially, every piece of content generated or processed by Claude will now carry an identifier indicating its AI-generated origin.

Why is a watermark being added to Claude's content?
There is growing concern regarding "AI slop"—low-quality AI-generated content—flooding the internet. Additionally, it is often difficult to determine whether specific content was created using AI.

Anthropic has updated its information regarding the marking of AI-generated content. According to the company, machine-readable marks are being embedded into content produced by the Claude model. This step has been taken to meet the transparency requirements mandated by Article 50 of the EU AI Act.

How will the invisible watermark work in text?
According to the details provided, an invisible watermark will be directly embedded into text generated by the Claude model. It will not be visible to the naked eye.
However, a key feature is that the mark can persist even if the text is copied and pasted into another document. It may also be difficult to remove through editing. Anthropic states that this marking process will not affect the meaning or quality of Claude's responses.

The system will differ for images and other files
The approach may vary—or be implemented differently—for files generated by Claude other than text. For file formats such as SVG, PNG, and JPG, Anthropic will attach signed provenance metadata. This metadata adheres to the open standard set by the Coalition for Content Provenance and Authenticity (C2PA). This standard is used to record information regarding the content's origin—or provenance.
The signed metadata can also indicate whether the files have been processed by Claude. Furthermore, it can reveal whether the file has been tampered with subsequently.

What is the user reaction to the watermark?
User reactions to this change have been mixed. Some are unhappy with the decision to add a watermark to every output generated by Claude, while others have expressed varied opinions online. Some users have even stated that they are considering cancelling their Claude subscription following this change, with a few describing it as the company's worst decision to date.
On the other hand, some users have supported the move, even taking a dig at those who do not disclose their use of AI in their content.

How can the watermark be detected?
Anthropic states that it is working on enabling users and third parties to verify Claude's embedded watermark and provenance metadata.
A detection system will be able to check whether a supported Claude mark is present in a given text or file. The company has said it will soon share more details regarding this detection mechanism.

Could there be limitations to this?
Yes... the company has clarified that the presence of the watermark does not, in itself, constitute conclusive proof that the entire content was generated by Claude.
This is because users might employ Claude for tasks other than content generation—such as proofreading, translation, summarization, or file conversion.
For instance, if you write content and submit the text to Claude solely for proofreading, the resulting output might still carry Claude's mark. Similarly, content processed by Claude can subsequently be edited, extracted, or combined with other material.

Which Claude models and services will this apply to?
According to the company, this marking will apply to the output of supported Claude models in regions worldwide where Claude is available.

This includes the Claude platform, Claude, Claude Code, Claude Cowork, and Claude Tags. Additionally, the system will apply to Claude services available through cloud partners such as AWS, Google Cloud, and Microsoft Foundry.

Meanwhile, work is currently underway to implement marking support for existing models launched prior to August 2, 2026; the regulations include a transition period for older models, and the company is working on adding marking support to them as well.

Disclaimer: This content has been sourced and edited from Amar Ujala. While we have made modifications for clarity and presentation, the original content belongs to its respective authors and website. We do not claim ownership of the content.