Defense in depth for autonomous AI agents

Microsoft Security Blog roundup (May 7–14, 2026): Multiple high-risk issues and defensive advances. Active exploitation of the newly disclosed Dirty Frag Linux local privilege‑escalation vulnerability is being monitored, with Microsoft Defender providing detections. Research highlights include Kazuar—a modular P2P nation‑state botnet attributed to Secret Blizzard—stealthy third‑party compromise intrusions, and exploitable misconfigurations and prompt‑injection flaws in AI agent frameworks that can lead to RCE and data leaks. Microsoft also published defensive innovations: a multi‑model agentic