Anthropic Embeds Machine‑Readable Watermarks in Claude to Meet EU AI Act
Anthropic’s support article confirms that every Claude model launched after Aug 2 2026 will embed an invisible watermark and signed provenance metadata in generated text and files. The marks are meant to satisfy the EU AI Act’s transparency code and can be detected by third‑party tools, though Anthropic notes they can be lost with heavy editing or copy‑paste.