Microsoft split responsibility for AI agents

🛡 Microsoft split responsibility for AI agents

The document “AI agent shared responsibility model” was published on Microsoft Learn: the Azure responsibility model is extended to autonomous AI agents for the first time — orchestration, tools, and memory layers are added, and zones are marked with a C/M/S matrix for SaaS, PaaS, and custom IaaS.

🌍 The customer is responsible for agent instructions, tool selection and permissions, memory isolation, and human confirmation of irreversible actions even in SaaS — this is a ready-made template for security policies, audits, and contracts when deploying agents on Azure.

👤 Building agents? This is a ready-made checklist: least privilege for each tool, re-authorization for actions, human confirmation for deletions and payments, memory isolation, step and budget limits. The principles apply to any agentic stack, not just Azure.

Source 1: https://learn.microsoft.com/en-us/azure/security/fundamentals/shared-responsibility-ai-agent