Anthropic adds Claude watermarks to meet EU transparency rules

Anthropic said text generated by Claude models launched on or after Aug. 2 will be watermarked, and that supported files will carry signed provenance metadata. The company said the change is meant to help it comply with European Union transparency rules on AI-generated content. The support page for Claude says the system uses two complementary techniques. Generated text gets an imperceptible embedded watermark that does not change meaning, quality or readability, while supported files receive signed provenance metadata attached to the file. Anthropic says the marking is applied at the model level, so it is present regardless of which Claude product or surface produces the output. The company said the marks apply wherever Claude is offered worldwide, including Claude Platform (API), Claude, Claude Code, Claude Cowork and Claude Tag, although some platforms or features may not support every marking type. For files, Anthropic says the provenance label follows the Coalition for Content Provenance and Authenticity, or C2PA, open standard. It can be added to supported file types such as .svg, .png and .jpg, and is meant to show that a file was processed by Claude and to help detect tampering. The company also said it is working to let users and third parties detect Claude’s embedded watermarks and provenance metadata. But it warned that detection is only a signal, not fully conclusive about provenance, because content can be marked even when Claude is used to proofread, translate, summarize or convert material, and marks may be missing if text is heavily edited, paraphrased, translated or too short. The move lands as Article 50 transparency obligations under the EU AI Act became applicable on 2 Aug. The European Commission says the related code of practice is a voluntary way to show compliance with those obligations, and that about 190 companies and organisations had signed it by the end of July.