Defense in depth for autonomous AI agents

Microsoft Security Blog posts highlight emerging AI-era risks and evolving threats: autonomous AI agents require updated defense-in-depth (application design, identity, human oversight) as prompt-injection flaws in agent frameworks can lead to RCE; cloud-native AI apps on Kubernetes are at risk from exploitable misconfigurations (exposed UIs, weak auth, risky defaults) that can enable RCE and data leaks; and Microsoft demonstrates AI-assisted detection tooling (synthetic attack logs, MDASH agentic security). Active threats include Kazuar, a modular P2P botnet attributed to Russian actor Secret