Anthropic quietly became the first major frontier AI lab to deploy production-scale invisible text watermarking across all its products — globally, with no opt-out — on August 2, 2026. Here is what the mechanism actually does, where it breaks down, and the one governance gap it does not close for enterprise operators.