Anthropic Adds Invisible Watermarks To Claude Text Under EU AI Act Rules

Anthropic is adding machine-readable marks to content generated by supported Claude models, including invisible text watermarks and signed metadata for files, as it implements transparency commitments linked to the European Union’s AI Act.

The markings will not be limited to Europe. Anthropic says supported Claude models will apply them wherever Claude is offered worldwide, including through its own products and major cloud platforms.

 

TL;DR

 

• Claude models launched on or after August 2, 2026 support machine-readable marking at launch.
• Generated text can carry an invisible watermark that may survive copying, pasting and some editing.
• Supported files can include signed C2PA provenance metadata.
• A detected Claude mark does not prove Claude originally authored the content.
• Anthropic is also working on detection tools for users and third parties.

 

How Will Anthropic Mark Claude-Generated Content?

 

Anthropic is using two methods to identify content generated or processed by supported Claude models.

The first is an imperceptible watermark embedded directly into generated text. Anthropic says the watermark does not alter the meaning, quality or readability of Claude’s response, but travels with the text when it is copied and pasted and may remain detectable after some editing.

Watermarking will be implemented at the model level, meaning it can appear regardless of whether the text was generated through Claude, Claude Code, Claude Cowork, Claude Tag or the Claude API.

Supported models accessed through AWS, Google Cloud and Microsoft Foundry will also carry embedded text watermarks.

For supported files such as SVG, PNG and JPG, Anthropic will use digitally signed provenance metadata based on the Coalition for Content Provenance and Authenticity, or C2PA, open standard.

The metadata can indicate that a file was processed by Claude and help determine whether the file has subsequently been tampered with. However, Anthropic notes that signed metadata may not be supported across every third-party platform.

“Greater transparency and signals about where content comes from can give people useful context about the information they consume,” Anthropic said in its guidance explaining the system.

 

What Does The EU AI Act Require?

 

The move comes as transparency requirements under Article 50 of the EU AI Act became applicable on August 2, 2026.

Under the rules, providers of relevant AI systems are required to add machine-readable marks that enable artificially generated or manipulated content to be detected. Deployers face separate disclosure requirements covering areas such as deepfakes and certain AI-generated text published on matters of public interest without human review or editorial control.

The European Commission says the requirements are intended to address risks including misinformation, manipulation, fraud, impersonation and consumer deception.

Anthropic’s marking system also forms part of its commitments under the EU’s Code of Practice on Transparency of AI-generated Content.

The code itself is voluntary, but the underlying Article 50 transparency requirements are legal obligations. By the end of July 2026, around 190 companies and organizations had signed the code.

 

Not necessarily.

Anthropic explicitly warns that detecting one of its marks only indicates that content may have been processed by Claude. Someone could, for example, submit human-written material to Claude for proofreading, translation or summarization, after which the resulting text could carry a Claude watermark.

Similarly, the absence of a detectable watermark does not prove that content was written by a human.

Anthropic says marks can become undetectable when text has been heavily edited, paraphrased, translated or combined with other writing. Very short passages may also contain too little text for reliable detection, while file metadata can disappear through screenshots, format conversions or re-saving.

Older Claude models are another limitation. Models launched on or after August 2, 2026 support marking at launch, while Anthropic says it is still working to introduce support for models released before that date.