Anthropic has announced that content generated by its Claude AI models will now include invisible, machine-readable watermarks and cryptographically signed provenance metadata to improve transparency and comply with new European Union regulations.
According to an updated support page published on Tuesday, the company said the watermarking system will cover text, text files and images generated by Claude.
The embedded watermarks are designed to be invisible to users and will not affect the quality, readability or meaning of AI-generated content.
Anthropic said text produced by any Claude model will automatically carry the watermark, which is expected to remain embedded even when users copy and paste the content elsewhere.
For supported file formats such as PNG, JPG and SVG, the company will also attach cryptographically signed provenance metadata that identifies whether the content was generated or processed using Claude and whether it has been altered.
The metadata will follow standards developed by the Coalition for Content Provenance and Authenticity (C2PA), enabling users and third parties to verify the origin and integrity of AI-generated content.
Anthropic said the new measures are intended to comply with Article 50 of the European Union’s AI Act, which requires providers of generative AI systems to clearly identify AI-generated content.
The company added that it is also developing AI detection tools that will enable users and organizations to verify embedded watermarks and provenance metadata independently.
