Threat modeling AI applications

Microsoft Security Blog (Feb 2026) highlights emergent risks and guidance across four themes: (1) developer-targeting supply chain attacks — a documented campaign used malicious Next.js repositories to trigger covert RCE-to-C2 chains through standard build workflows, demonstrating how staged command-and-control can hide in routine development tasks; (2) AI/agent-era risk management — guidance on threat modeling for probabilistic and agentic AI, and a Cyber Pulse report showing widespread agent adoption and the need for observability, governance, and security; (3) agent/runtime risks — warnings