Anthropic's Claude will invisibly watermark AI-generated text: How it works

Anthropic says new Claude models will embed machine-readable watermarks in generated text that can survive copy-pasting and some editing

Anthropic is adding invisible watermarks to text generated by Claude, allowing users and third parties to identify whether a piece of text has been processed by its AI models even after it has been copied and pasted elsewhere.

 

The company detailed the system in an updated Claude Help Center article, saying Claude models launched on or after August 2, 2026 will support machine-readable marking at launch. Generated text will carry embedded watermarks, while supported generated files will include digitally signed provenance metadata.

Anthropic said the move is part of its commitments under the European Union's AI Act's Article 50(2) Code of Practice on Transparency of AI-Generated Content. The company said the marking system will apply to supported Claude models across its products and services, including Claude, the API, Claude Code, Claude Cowork and Claude Tag, as well as wherever Claude is offered worldwide.

How Claude's invisible watermark works

The watermark is embedded directly into the generated text and is designed to be imperceptible to readers. Anthropic said it does not change the meaning, quality or readability of Claude's responses.

 

Because the mark is embedded in the text itself, it can travel with the text when it is copied and pasted elsewhere and may survive some editing. Anthropic said the watermarking happens at the model level, meaning it is not tied to a particular Claude product or interface.

The system is separate from the provenance metadata that Anthropic plans to attach to supported files. For formats such as SVG, PNG and JPG, Claude will add digitally signed provenance metadata based on the Coalition for Content Provenance and Authenticity (C2PA) standard. This can indicate that a file was processed by Claude and help determine whether the metadata has been tampered with.

 

Anthropic said embedded watermarks will also apply when supported Claude models are accessed through AWS, Google Cloud and Microsoft Foundry, although signed provenance metadata may not be available on every platform.

Claude watermark will not prove AI authorship

The marking system is not designed to work as a definitive AI detector. Anthropic itself cautioned that finding a Claude mark only indicates that content may have been processed by Claude.

 

For example, a user could provide their own writing to Claude for proofreading, translation or summarisation. The resulting text could carry a Claude watermark even though the underlying ideas and material came from the user or another source.

 

The absence of a watermark also does not prove that content was written without AI. Anthropic said marks may not be detectable if content was generated by an older model, heavily edited, paraphrased, translated or mixed with other writing. Very short passages may also not contain enough text for a reliable signal.

 

This means the system is intended as a provenance signal rather than a standalone AI-content detector.

 

Anthropic said it is working on tools that will allow users and third parties to detect Claude's embedded watermarks and provenance metadata. 

Older Claude models will also get marking

The watermarking system will not initially cover every Claude model. Anthropic said models launched before August 2, 2026 are covered by a transition period under the EU rules, and the company is working to add marking support to those models as well.