Claude to watermark AI productions for Europe
Anthropic, the maker of AI system Claude, is going to add an invisible watermark to everything produced with its new models.
This will be done not just for Europe but worldwide, and is intended to make it clear whether something has been created with AI, and if so, what. In doing so, the company is submitting itself to a specific article in Europe’s AI Act that obliges model developers to do this.
The aim is to provide transparency around responsibility for AI-generated output. This is divided into four categories, which also include company helpdesks and, for example, news publishers.
“Because the watermark is part of the text, it will travel with the text when it is copied and pasted elsewhere, and it can also be preserved after edits.“
Watermarking is being introduced in all new Claude models that have been in use since 2 August 2026. It does not yet apply to Fable 5, Opus 5 and Opus 4.8. That will follow.
This concept is used in the physical world, for example, to incorporate security features into banknotes. It helps distinguish genuine from fake. The features are not perceptible to the naked eye, however, machines can see them.
The metadata is applied according to the open standard of the Coalition for Content Provenance and Authenticity (C2PA).
Claude users are not universally enthusiastic about it.
The English tech title The Register cites various experts and situations in which these kinds of protection systems are bypassed or destroyed.
For now, Anthropic is the first to submit itself to the AI Act in this way.
Emerce
Subscribers 0
Fans 0
Followers 0
Followers