Anthropic has announced it will start embedding invisible, machine-readable watermarks in text and images generated by its Claude AI models, in order to comply with new EU transparency rules. The move, disclosed on a new Claude support page, means generated text will carry embedded watermarks while generated files will include digitally signed provenance metadata where supported, making it easier for people and platforms to identify AI-generated content without altering how it looks or reads.

The changes stem from the EU's AI Act, which took effect on 2 August 2026 and includes a four-month grace period for existing products, meaning new Claude models will be watermarked from launch while support for existing models is still being rolled out. The marks will apply across Claude Platform, Claude, Claude Code, Claude Cowork and Claude Tag, using C2PA provenance metadata for images (a standard also used by Adobe, OpenAI and Google) and an unnamed "imperceptible" text watermark that persists through copying and some editing. Anthropic says it will publish further technical details on how users and third parties can detect these marks.