Defense in depth for autonomous AI agents

This Microsoft Security Blog feed aggregates multiple high-priority posts covering AI security, nation-state malware, cloud misconfigurations, and an active Linux kernel escalation: 1) Guidance on defense-in-depth for autonomous AI agents emphasizing application-layer design, strong identity, and human oversight; 2) Analysis of Kazuar, a modular P2P botnet attributed to Russian actor Secret Blizzard; 3) Warnings that exploitable misconfigurations in cloud-native AI apps (exposed UIs, weak auth, risky defaults) can lead to RCE and data leaks and that prompt-injection in agent frameworks can be‑