Anthropic adds invisible watermarks to Claude AI text, helping identify AI-generated content while meeting new EU transparency rules under the AI Act.
Anthropic, the company behind the Claude AI chatbot, has started adding hidden watermarks to text that Claude creates. The watermarks cannot be seen by readers, but computers can detect them. This move is meant to help people know when a piece of writing came from an AI system. The change is already rolling out and will affect Claude users all over the world.
The new watermarks are built directly into the words Claude generates. They do not change the meaning, tone, or quality of the text in any way. Anthropic says the marks are designed to stay with the text even after it is copied and pasted somewhere else. In many cases, the watermark can also survive some editing by a human.
This step is connected to new rules in the European Union. The EU’s AI Act requires companies to make AI-generated content easier to identify. Anthropic said it is following a part of this law called Article 50, which focuses on transparency. Even though the rule comes from Europe, Anthropic is applying the watermark to Claude everywhere, not just in the EU.
The watermarking system covers many Claude products. This includes the Claude app, the Claude API, Claude Code, Claude Cowork, and Claude Tag. It also applies when Claude is used through other companies, such as Amazon Web Services, Google Cloud, or Microsoft. Only newer Claude models, launched on or after August 2, 2026, support this feature right away.
Anthropic explained that older Claude models will get similar watermarking abilities later. The company is still working on adding this feature to those earlier versions. It also plans to release tools that let people check if a piece of text carries a Claude watermark. These detection tools have not been shared publicly yet.
Anthropic was careful to note that the watermark is not perfect proof of who wrote something. A person might use Claude to edit or polish text they wrote themselves, and that text could still carry a mark. On the other hand, text with no watermark does not always mean a human wrote it, since heavy editing or short passages can weaken or remove the signal.
In short, Anthropic is now hiding invisible markers inside Claude’s writing to support honesty about AI-made content. The change follows new European rules, but it will apply everywhere Claude is used. More tools and details, including ways to detect these watermarks, are expected in the coming months.
Trending
