Anthropic has announced plans to embed machine-readable watermarks and provenance metadata into content generated by its Claude AI models, citing compliance with European AI transparency rules as the primary driver.
The commitment, published on a new Claude support page in August 2026, covers both text and images. Generated text will carry imperceptible watermarks woven directly into the output, while image files will include digitally signed provenance metadata using C2PA — a standard already adopted by Adobe, OpenAI, and Google. Anthropic says the marks are invisible to human readers but will allow people and online platforms to detect Claude-generated content.
The changes are not yet in effect. The EU’s AI Act, which introduced new AI labeling and transparency obligations on August 2nd, 2026, includes a four-month compliance grace period for AI products that launched before that date. Anthropic says new Claude models will apply the markings from day one upon release, while support for existing models is still being developed.
The watermarking will apply globally across Claude Platform (API), Claude, Claude Code, Claude Cowork, and Claude Tag, and will also be present when Claude models are accessed through AWS, Google Cloud, or Microsoft Foundry. Because the text watermark is embedded in the output itself, Anthropic says it “will travel with the text when it’s copied and pasted elsewhere, and may persist through some editing.”
Anthropic says it is working on tools that would allow users and third parties to detect these marks, with technical documentation to follow. However, the company acknowledges the systems are not foolproof, noting that content lacking detectable marks could still originate from generative AI. C2PA metadata is also known to be easily stripped, sometimes unintentionally when files are uploaded to online platforms, raising questions about the reliability of the image-side approach.
Source: The Verge