Anthropic says new Claude models will mark generated and processed content

Anthropic said it will roll out machine-readable watermarks for content generated or processed by its models, according to Ars Technica. The company said new models offered globally will mark AI-generated content from their release, while generated files will carry digitally signed provenance metadata where supported. The report says Anthropic intends to apply marks to processed content as well, including uses that may be standard editing rather than substantial alteration. Anthropic cited compliance with the EU AI Act, which requires markings for AI-generated or manipulated outputs and provides a grace period through December 2026 for previously released models.