Microsoft Azure's new access control stops AI agents from going rogue

Microsoft researchers have published a paper detailing a new decentralized, multi-layered access control architecture designed specifically for agentic AI systems operating in critical cloud infrastructure. Deployed in production at Azure across hundreds of datacenters, the system addresses the fundamental security challenge that traditional role-based access control (RBAC) cannot handle: AI agents exhibit stochastic, non-deterministic behavior, making conventional trust models insufficient.

The framework introduces four key innovations: a compound identity model that binds every agent action to delegated human authority, a hierarchical permission system spanning five granularity levels from global platform access down to per-parameter constraints, a decentralized policy ownership model where tool teams independently govern their authorization boundaries, and progressive trust escalation with safety interlocks that prevent autonomous agents from executing high-risk operations without additional verification.

In production, the system governs 20+ specialized AI agents and 60+ deterministic playbooks processing thousands of operations daily. Empirical results show zero unauthorized write operations over eight months, with detailed data on access pattern distributions, denial rates, and the effectiveness of layered authorization in preventing privilege escalation by non-deterministic actors. The design is grounded in the OWASP Top 10 for LLM Applications (2025) threat taxonomy, with each architectural decision mapped to specific attack vectors.