Anthropic announced that its Claude models will begin to incorporate invisible watermarks in the texts they generate, a measure that responds to the transparency laws of the European Union. The functionality will apply to versions available worldwide.

According to the available information, these marks do not alter the content visually, but are integrated directly into the text generated by the model. Their objective is to allow the identification of AI-generated content without affecting the user's reading.

The company has not detailed how these marks will be detected or what tools will be used to verify them. This has generated discussion in the community about the effectiveness and transparency of the labeling mechanism.

The move comes at a time when the EU is pushing for strict regulations on the labeling of AI-generated content, requiring providers to implement traceable identification systems that make it possible to distinguish AI-generated content from human-created content.