cyber.hendryadrian.com
Security News · securityweek.com

Microsoft Adds Risk Scoring and Dynamic Permissions for Enterprise AI Agents

· Original article ↗

Summary

Microsoft announced Project Perception controls that assess enterprise AI agents’ risk, grant task-specific permissions, and can restrict or halt agents when risk rises.

Key points

  • Dynamic Permissions starts agents with no or minimal access, grants permissions per task, and lets them expire or shrink as tasks change or end.
  • High-impact actions can require human review.
  • Dynamic Agent Risk continuously scores agents using activity and identity, data, endpoint, cloud, network, and runtime signals.
  • Dynamic Conditional Access can reduce permissions, limit data or network access, require approval, isolate an agent, or halt execution as risk changes.
  • Project Perception Playbooks define response goals; Guardian Agents investigate risk changes and can suggest or enforce containment under human-set rules.
  • Microsoft says discovery is expanding to 60 types of local agents, including Claude Code, OpenAI Codex, and GitHub Copilot.

Article Details

Event Type
Microsoft announced new security controls for monitoring and restricting AI agents in enterprise environments.
Impact
The controls aim to reduce risks from overprivileged, manipulated, compromised, or misbehaving agents through task-based permissions, continuous risk scoring, conditional access, human approval, and runtime isolation or halting. Microsoft also expanded agent discovery to 60 supported types.

People

Hayete GallotMicrosoft Security executive vice president who described risks from organizations trusting deployed agents by default.Nadim AbdoMicrosoft identity and network access corporate vice president who clarified that the insider framing does not mean agents are inherently malicious.

Vendors

MicrosoftMicrosoft on Monday announced new controls to monitor and restrict potentially rogue AI agents in enterprise environments, arguing that organizations should treat these agents as insiders rather than trusted software.

Products

Claude CodeMicrosoft is also expanding discovery to 35 more types of local agents, bringing the supported total to 60, including Claude Code, OpenAI Codex and GitHub Copilot.Dynamic Agent RiskDynamic Agent Risk gives each agent a risk score that is continuously updated, drawing on the agent’s activity as well as identity, data, endpoint, cloud, network and runtime signals. The score factors in the agent’sDynamic Conditional AccessDynamic Conditional Access, which brings Entra Conditional Access to agents, acts on those risk changes.Dynamic PermissionsOne of the new features, Dynamic Permissions, gives agents no permissions, or the bare minimum, at the outset.Entra Conditional AccessDynamic Conditional Access, which brings Entra Conditional Access to agents, acts on those risk changes.GitHub CopilotMicrosoft is also expanding discovery to 35 more types of local agents, bringing the supported total to 60, including Claude Code, OpenAI Codex and GitHub Copilot.OpenAI CodexMicrosoft is also expanding discovery to 35 more types of local agents, bringing the supported total to 60, including Claude Code, OpenAI Codex and GitHub Copilot.Project PerceptionThe capabilities extend Project Perception, the agentic security system Microsoft launched in July.

Related Articles