Anthropic is introducing new ways to identify content generated or processed by Claude, with the company planning to add machine-readable marks to AI-generated text and files. The move comes as Anthropic signs the EU AI Act’s Article 50(2) Code of Practice on Transparency of AI-Generated Content.

Under the company’s new approach, Claude models launched in the EU on or after August 2, 2026, will support content marking from day one. Anthropic also plans to bring the technology to older models, although those models are covered by a transition period under the law.

Two different ways to identify AI-generated content

The marking system will use two different methods depending on the type of content Claude produces. For text, Anthropic will embed an invisible watermark directly into the generated content. The company says users will not be able to see the watermark, and it will not affect the meaning, readability, or quality of the text. Since the mark is embedded into the text itself, it can also remain when the content is copied and pasted elsewhere, although heavy editing can affect it.

For supported files, Claude will use signed provenance metadata instead. This will follow the C2PA standard, which is designed to provide information about where digital content came from. When supported, files such as PNG, JPG, and SVG images will include this metadata, allowing users to check whether the file was processed by Claude and whether the metadata indicates that it has been altered.

The markings will apply across supported Claude products, including the Claude API, Claude, Claude Code, Claude Cowork, and Claude Tag. Anthropic also says the system will work worldwide, rather than being limited to users in the EU. Supported Claude models accessed through AWS, Google Cloud, and Microsoft Foundry will also receive embedded text watermarks, although signed file metadata may depend on the platform.

Anthropic is also working on tools that will allow users and third parties to detect these marks. More information about how detection will work is expected in future technical documentation.

However, Anthropic is making it clear that these marks will not provide definitive proof of where content came from. A detected mark only indicates that the content may have been processed by Claude. For example, someone could use Claude to translate, summarize, or proofread content that originally came from somewhere else.

About the Author

Abdul Haddi

Abdul is a tech writer and Editor for PC Guide, specializing in all things tech, gaming, and hardware.