Claude has started secretly embedding invisible watermarks into AI-generated text to make the text easier to identify and trace in a bid to ensure transparency and academic integrity. 

The move has sparked serious backlash among users who fear their writing could now be traced back to the chatbot.

Claude chatbot’s parent-company, Anthropic, introduced the new watermarking system in its latest AI models released after August 2, 2026, meaning every piece of text generated by supported versions of Claude carries an invisible machine-readable mark.

The watermark cannot be seen by users and does not change the appearance or meaning of the text. However, Anthropic says it is designed to remain embedded in the text even after it is copied and pasted or undergoes minor edits.

The company says the technology will eventually allow users and third parties to determine whether a piece of writing was generated by Claude. Anthropic is also working on detection tools that could identify the hidden marks.

The development has already triggered backlash among Claude users, particularly those who use AI to help write books, academic work, professional documents and other forms of content.

Some users have raised concerns that the watermark could expose their use of AI even when they have made significant contributions themselves. Others questioned whether people should be able to use Claude without having every piece of text secretly marked by the company.

Anthropic has defended the move as part of its effort to make AI-generated content easier to identify and trace. 

The company is also trying to comply with new European Union rules requiring AI-generated content to be marked in a machine-readable way.