Artificial intelligence is becoming a bigger part of everyday life, but governments are also introducing new rules to make AI-generated content easier to identify. In response to upcoming European Union requirements, Anthropic has announced a new invisible watermarking system for text produced or processed by its Claude AI chatbot.
%20(1).jpg "Anthropic Unveils AI Watermarks to Comply With EU Law")
The American AI company introduced the feature in San Francisco, explaining that the watermark will help meet new EU regulations requiring AI-generated content to include machine-readable markings. Anthropic said the system will be used globally, rather than being limited to European users.
Unlike a visible label or symbol, the new Claude watermark cannot be seen by people reading the text. It is designed to remain with the content even when users copy and paste it into another document or platform.
However, Anthropic stressed that the presence of a watermark does not automatically prove that Claude originally created the content.
For example, someone might write an article themselves and then use Claude only to correct grammar, translate it, summarize it, or convert it into another format. The resulting text could still contain a Claude watermark even though the original ideas and information came from the user.
The system also has limitations.
The system also has limitations. Very short text or content that has been heavily edited may not contain the watermark. This means the technology should be treated as an indicator that Claude processed content rather than definitive proof of its origin.
Anthropic is also introducing provenance information for supported files. When Claude creates a file such as a JPG image, the system can attach digitally signed metadata based on an open standard designed to record the history and origin of digital content.
Such technology could become increasingly important as AI-generated images, documents, videos and other media become more common online. Digital provenance can help platforms and users understand how content was created or modified.
Anthropic said models launched on or after August 2 will include the new watermarking technology, while older Claude models will also be upgraded.
Anthropic's move is part of a broader effort by technology companies to improve transparency around AI-generated content. Around 190 organizations have signed a voluntary European Union code of good practice for AI, including major companies such as Google, Meta, Microsoft, Mistral and OpenAI.
Google has already developed SynthID, a digital watermarking and detection technology designed to identify AI-generated content. Google is also encouraging other major technology companies to support common standards for AI content identification.
Anthropic, however, has not yet released a dedicated tool for detecting its Claude watermarks, despite the EU code calling for such capabilities.
Anthropic's new system highlights how quickly AI regulation is changing. As AI tools become capable of producing increasingly realistic text and media, governments and technology companies are looking for reliable ways to distinguish AI involvement from human-created content.
The Claude watermark will not solve every problem, but it represents another step toward greater transparency. As AI adoption continues to grow, similar systems could become a normal part of digital content creation, helping users, businesses and regulators better understand where online content comes from.
