Microsoft gives AI agents access that shrinks with risk
Microsoft announced dynamic permissions and risk controls for Project Perception, plus discovery covering 60 agent types, extending its system to agents running inside organizations. Project Perception also shares an agent registry and signals with Microsoft Agent 365.
Agents can start with no or minimal access, then receive permissions for a task that shrink or expire when it changes or ends. Continuous risk scoring can trigger read-only access, human approval, or a stop; Guardian Agents investigate risk changes and execute responses set in Playbooks. These controls can narrow an agent’s reach before broad access persists unnoticed.
Yet discovery and automated response do not assign ownership. A separate Splunk survey found 38% of 2,050 IT and engineering respondents lacked clearly defined accountability for AI monitoring. If employee-built workflows outlive their creators, dynamic permissions may limit their reach while responsibility for their credentials and actions remains unclear; Microsoft has not shown how reliably the announced controls work.
AI Agent Adoption Creates Risk as Oversight Struggles to Keep Pace
Rapid adoption of agentic AI comes with potential business benefits. But if done incorrectly it also carries risks around AI activity management and security
Microsoft Targets Rogue AI Agents With Dynamic Permissions, Risk Scoring
Microsoft says companies should treat AI agents as insiders, and is adding controls that grant access per task and tighten it as an agent’s risk rises.