Anthropic announced on Tuesday that its Claude AI platform will begin watermarking all artificially generated content to comply with European Union regulations. The company has adopted the EU AI Act’s transparency requirements and plans to implement these standards across its operations.
Beginning August 2, 2026, all newly launched Claude models will include watermarking features from the start. Text-based outputs will contain embedded watermarks that persist when content is copied and pasted, while image files and other documents will incorporate digitally signed metadata using the Coalition for Content Provenance and Authenticity standard. Despite the EU focus, these watermarks will be deployed globally wherever Claude services operate, though some platforms may have technical limitations.
Anthropic acknowledged that watermarking technology has inherent limitations. Content that undergoes significant editing, paraphrasing, or translation may lose detectable watermarks, and file metadata can be removed through screenshots or format conversions. The company plans to release detailed guidance on identifying watermarks in future documentation and is working to add watermarking capabilities to existing models that predate the August 2026 deadline.
Anthropic joins other major artificial intelligence companies including OpenAI, Google, and Meta in committing to transparency measures for machine-generated content.
Originally reported by Mashable →
About the Author
