Defense in depth for autonomous AI agents

A set of Microsoft Security Blog posts (May 2026) covering emerging AI-security risks, active exploitation, and defensive advances. Key items: Dirty Frag, a Linux kernel networking/memory-fragment vulnerability enabling reliable local privilege escalation post‑compromise; prompt‑injection and other RCE vulnerabilities in AI agent frameworks and exploit-prone misconfigurations in cloud-native AI apps; Kazuar, an evolving P2P nation‑state botnet attributed to Secret Blizzard; and stealthy intrusions via trusted third‑party compromise. Defensive content includes Microsoft Defender detection/tele­