The company Anthropic will start labeling content from its Claude language models. The new rules will affect all text and image files of users. The labeling was introduced due to the new requirements of the European Union. The tags will work for users all over the world. New Claude models will receive a labeling system immediately after launch. The models will start adding tags later until the second of August. The generated text will contain a special invisible watermark. This character is saved when copying text between applications. Graphic files will receive digitally signed metadata. The system supports the popular .svg and .png image formats. Labeling also works for .jpg files. All markings comply with the open C2PA standard. This standard adds information about the file source. The standard does not support tables yet.xlsx and presentations .pptx.

The law on content labeling came into force on August 2. The European Union requires that all content generated by artificial intelligence be marked. The company has decided to extend the labeling to all regions. The models will tag the content regardless of the user's country. The company has not yet disclosed the technical details of the system. Experts do not know the exact methods of protection against deletion of marks.

The Register considers the removal of markings to be a fairly simple task. If desired, users will be able to clear files of labels. The company has kept the ways of marking protection a secret. Industry experts continue to study the new requirements of the European Union.