From SiliconANGLE · Duncan Riley · · 1m 40s
Microsoft targets 'agentic insider' with AI agent controls
Microsoft adds security controls to stop rogue AI agents from acting without human oversight.
In brief
Microsoft adds security controls to stop rogue AI agents from acting without human oversight. Microsoft introduced new security features to limit AI agents' actions based on risk scores, preventing unauthorized behavior. Originally reported by SiliconANGLE.
Microsoft introduces security controls
Microsoft Corp. today introduced security controls for rogue artificial intelligence agents that cut back what an agent may do as its assessed risk climbs.
Hayete Gallot identifies 'agentic insider' risk
Hayete Gallot, executive vice president of Microsoft Security, called the 'agentic insider' in a blog post, arguing enterprise agents are often overprivileged and frontier models cannot be fully trusted.
Satya Nadella emphasizes trust minimization
Satya Nadella, Microsoft Chairman and Chief Executive, urged companies to treat models as insider risks with safeguards kept outside the model itself, stating the most trustworthy system enables trust the least.
Dynamic Agent Risk tracks risk scores
Microsoft’s Dynamic Agent Risk feature continuously updates an agent’s risk score based on signals from identity, endpoint, cloud, and network tools, climbing when behavior drifts from intended use.
Dynamic Conditional Access adjusts permissions
Dynamic Conditional Access decides actions as risk scores rise, allowing agents to lose write access, require human approval for high-impact tasks, or isolate and stop runtime entirely.
Dynamic Permissions limit agent authority
Dynamic Permissions grant agents temporary access to what a single task needs, with companywide limits preventing agents from holding more authority than their human user.
Project Perception gets new security tools
Project Perception, Microsoft’s agentic security system, now includes playbooks for security staff to set outcomes and Guardian Agents that can recommend or apply containment based on risk score shifts.
The bottom line
Microsoft introduced new security features to limit AI agents' actions based on risk scores, preventing unauthorized behavior.
