Anthropic embeds watermarks in Claude-generated text and images for EU AI Act compliance

According to TechCrunch and The Verge, Anthropic announced plans to embed machine-readable watermarks into Claude-generated text and digitally signed provenance metadata into generated files to comply with European AI transparency rules. The move operationalizes EU AI Act requirements for identifying artificially generated content.

Update (2026-08-18): Anthropic released technical details on Claude's watermarking system, which uses Google DeepMind's open-source SynthID-Text technology to embed imperceptible markers in AI-generated text for European AI Act compliance. According to TechCrunch, Anthropic explained that watermarking works by ...