Claude's AI text now carries an invisible watermark
What's the story
Anthropic has announced its plan to add invisible watermarks to its new Claude models. The move is in line with the transparency requirements of the European Union's AI Act, Article 50. The company said that "Generated text will carry embedded watermarks, and generated files will include digitally signed provenance metadata where supported." Claude models deployed on or after August 2, 2026, feature native machine-readable marking. Older models are set to receive the capability in subsequent updates.
Feature details
Watermarks will be invisible to human readers
The watermarks will be embedded directly into the text generated by Claude models, making them invisible to human readers.
However, machines will be able to detect these marks if the text is copied and pasted from Claude.
Some watermarks may even survive editing, according to Anthropic.
The company has also clarified that this feature will not be limited to European users but applied globally across all platforms supporting its models.
Output support
Support across other outputs
The watermark feature will be supported across other outputs like API, Claude, Claude Code, Claude Cowork, and Claude Tag.
This means cloud customers will receive marked outputs if they access supported models through major platforms such as AWS, Google Cloud, and Microsoft Foundry.
Notably, files generated using Claude models will also carry signed provenance metadata, according to the Coalition for Content Provenance and Authenticity (C2PA) standards.
Impact assessment
Revolutionizing AI detection
The introduction of invisible watermarks by Anthropic could revolutionize AI detection. It means that AI-generated text can now be traced on a large scale, something that was not possible before.
This could apply to anything from school essays and corporate reports to news articles, research papers, software documentation, social media posts, and legal documents.
However, it's worth noting that heavy paraphrasing can remove existing watermarks from texts.
Sector influence
Impact on education and social media
The biggest impact of this new approach by Anthropic is likely to be on education and social media.
It will be easier to verify if content on social media was generated by AI.
In the field of education, schools and universities will have a way to determine with some degree of certainty whether a student's submission was generated by AI or not.