How agentic AI flips the trust model

### Executive Summary: This article discusses how agentic AI and LLMs are changing AppSec by moving risk into the data and context layer—introducing new attack vectors like prompt injection and poisoned context windows—and outlines mitigation strategies (data provenance, least privilege, containment, runtime monitoring, egress controls, and governance gates) and organizational approaches (tokenization, confidential computing, and supply-chain assessments) to manage emerging "trust debt".