Claude now embeds invisible watermarks in AI-generated text

Anthropic, the creators of Claude, have introduced a new feature that embeds invisible watermarks into the text generated by their AI model. This hidden signal, machine-readable and lasting through copying and some editing, aims to enhance transparency in AI-generated content, in line with the European Union's AI Act Code of Practice on Transparency of AI-Generated Content.

Claude models launched on or after August 2, 2026, incorporate this marking from their initial launch, though older models are also being updated. The watermark is embedded directly into the generated output, ensuring it travels with the text when users copy and paste. While Anthropic is working on detection systems, they caution that detecting a Claude mark is not definitive proof of AI authorship, as people often use Claude for various tasks.

Conversely, the absence of a detectable mark does not guarantee human authorship. These watermarks can be lost through heavy editing, paraphrasing, translation, or mixing with other texts. Very short passages may also lack sufficient material for reliable detection. This development arrives at a time when AI-generated content is increasingly prevalent, leading to concerns about its authenticity and the potential for AI to assist or replace human work.

In educational settings, such watermarks could help institutions detect AI-assisted submissions, aligning with the Nigerian government's initiative to incorporate AI into education while maintaining human oversight.