Anthropic has announced that it will embed an 'invisible watermark' into text generated by Claude, and this will be applied worldwide, including in Japan.

AI company Anthropic has announced plans to embed 'invisible watermarks' into text generated by its AI chatbot 'Claude.' Images and other files will be given C2PA-compliant metadata indicating the content's origin, allowing the system to automatically identify content generated and processed by Claude.

How Claude marks AI-generated content | Claude Help Center
https://support.claude.com/en/articles/16266773-how-claude-marks-ai-generated-content

Anthropic says it will watermark text generated by its AI models | TechCrunch
https://techcrunch.com/2026/08/11/anthropic-says-it-will-watermark-text-generated-by-its-ai-models/

Claude will apply invisible watermarks to AI text and images | The Verge
https://www.theverge.com/ai-artificial-intelligence/977823/anthropic-claude-ai-watermarks-c2pa-text-images

Anthropic has signed the EU's Code of Conduct on Transparency for AI-Generated Content under its AI Law, and as part of this effort, it has announced plans to mark Claude-generated content with machine-readable watermarks. While compliance with EU regulations is the impetus for this change, the watermarks will not be limited to the EU, but will apply to all regions worldwide where the corresponding Claude model is available, including Japan and the United States.

New Claude models released in the EU after August 2, 2026, will support watermarking from the time of release. Anthropic also plans to gradually add watermarking support to existing models released before that date.

Watermarking will be applied not only to Claude's web applications, but also to various services that utilize compatible models, such as Claude Platform (API), Claude Code, Claude Cowork, and Claude Tag. Furthermore, even when using Claude through AWS, Google Cloud, or Microsoft Foundry, watermarks will be embedded in the text generated by compatible models.

Regarding the text, Claude will embed an 'imperceptible watermark' into the text itself that it generates. According to Anthropic, this watermark is embedded without changing the meaning, quality, or readability of the text, and will move with the text when it is copied and pasted elsewhere, and may remain even after some editing. At the time of writing, the specific technology used to embed the watermark has not been revealed.

On the other hand, supported files such as PNG, JPEG, and SVG generated or processed by Claude are accompanied by digitally signed provenance information. This metadata uses an open standard developed by the Coalition for Content Provenance and Authenticity (C2PA), which includes Adobe and Google, allowing users to verify that a file was processed by Claude and whether it has been modified since processing.

Anthropic also plans to provide a mechanism that will allow users and third parties to detect Claude's watermarks and provenance information. Specific detection methods will be explained in a technical document to be released in the future.

However, Anthropic cautions that the presence of a watermark does not necessarily mean that 'Claude created the content from the beginning.' For example, a watermark may be added to human-written text that has been proofread, translated, or summarized by Claude, and parts of the text may have been edited or combined with other text afterward. In other words, a watermark indicates the 'possibility that the content was processed by Claude,' and does not prove the complete origin of the content.

Conversely, the absence of a watermark does not necessarily mean that the content was not generated by AI. Watermarks may not be detected if the text has been significantly edited, rephrased, or translated; if the text is too short to be detected; or if the content was generated using a Claude model prior to the implementation of watermarking. Similarly, C2PA metadata may be lost in files due to format conversion, resave, screenshots, etc.

TechCrunch points out that, in response to the EU's transparency regulations on AI-generated content, AI companies and content platforms are increasingly introducing mechanisms to identify AI-generated content. Anthropic's mechanism will be positioned not as a complete AI identification method, but rather as additional information to determine the origin of content generated and processed by AI.