Anthropic to watermark AI-processed content in all new models

Anthropic has announced that it will add machine-readable watermarks to content processed, not just generated, by its models, releasing them with all new models globally from day one. The move is to comply with the European Union's AI Act, which mandates watermarking of AI-generated or manipulated content, with a deadline for new models after August 2 and a grace period until December 2026 for previously released models. Text outputs will have invisible embedded watermarks, while other generated files will include digitally signed provenance metadata where supported. Interestingly, Anthropic is applying watermarks to all processed content, including cases where the AI only performs an assistive function like grammar correction, which the EU guidance does not require. This may result in stamping content that the law intended to exclude, and the true extent of watermarking will be unclear until Anthropic releases a detection tool, which it plans to do to offer technical support required by the EU law. For non-text content, Anthropic will use the C2PA metadata approach, though watermarks may not work on some platforms or features.