Anthropic adds persistent watermarks to AI-generated text and files
Anthropic has announced that it will implement invisible digital watermarks on text and files generated by its AI models, including Claude. The move comes in response to the European Union AI Act's Transparency Code, which came into effect on August 2 and mandates clear identification of AI-generated or edited content. Watermarking will be applied at the core model level across all interfaces, including the Claude platform API, enterprise tools, and cloud provider integrations. Embedded directly into generated responses, these machine-readable text marks are designed to travel with the content during copy-and-paste actions and withstand light editing. For exported media files such as images, Anthropic is adopting the open C2PA standard to attach cryptographically signed provenance metadata. While all models released after August 2 will feature default watermarking, support will eventually be retrofitted to legacy architectures. Anthropic joins other industry leaders like OpenAI, Google, and Meta in committing to the EU's strict content tracking rules.