Zero-Trust AI Architecture: Securing Autonomous Agents, APIs, and Decision Systems

Traditional zero-trust models were designed for human-initiated network traffic. AI systems introduce fundamentally new attack surfaces: autonomous agents making consequential decisions, API chains that execute without human intervention, and machine identities with asymmetric capabilities. This report extends zero-trust principles to cover AI-specific threats, defines microsegmentation for agentic systems, establishes continuous verification protocols for machine identities, and implements least-privilege frameworks that govern what autonomous systems can do.