Zero Trust for AI Is a Trust Boundary Problem
AI incidents that look model-driven often trace back to information trust boundaries that were never designed for high-scale, cross-domain consumption. Treating AI as a consumer of enterprise data reframes hallucination, inconsistency, and leakage as symptoms of upstream access, provenance, and auditability gaps. The article examines confidentiality as a boundary problem, integrity as a provenance problem, and decision risk as an operating model problem. It contrasts a technology upgrade posture with a system redesign posture to show second-order consequences for governance and accountability. A short vignette illustrates how ordinary workflows become unsafe when AI removes human friction and recombines context.