Microsoft is expanding its AI security strategy to address risks from autonomous AI agents. These agents can access company data, use applications and perform tasks with limited human input. The company announced new protections on October 19, 2026. The update focuses on monitoring agent activity and controlling access to business systems.
Microsoft calls this emerging threat the “agentic insider” risk. An AI agent can create security problems if it behaves unexpectedly or falls into the wrong hands. Companies must therefore monitor these systems and limit what they can access.
How Microsoft plans to secure AI agents
Microsoft is expanding Project Perception to help security teams monitor AI agents. The system uses Guardian Agents to assess activity and identify potential risks.
Security teams can use these tools to investigate suspicious behaviour and respond to emerging threats. Microsoft also plans to use security playbooks to coordinate protective actions. These workflows can help teams investigate incidents and contain potential damage.
Microsoft says human operators will continue to define security policies and approval rules. Guardian Agents will monitor activity and respond within those boundaries.
AI agents will receive task-based permissions
Microsoft is also focusing on Dynamic Permissions. This approach gives an AI agent access only to the resources it needs for a specific task.
Companies can restrict permissions by task and time. They can also remove access when an agent finishes its work. Sensitive actions may require human approval.
These controls aim to reduce the damage caused by compromised or misbehaving agents. Microsoft wants security teams to manage access across identity systems, data, devices, networks and cloud services.
Why enterprise AI security matters
Businesses are increasingly exploring AI agents to automate routine work. These systems can handle tasks across applications and business platforms. However, broader access also creates new security challenges.
Traditional security measures often focus on employees, devices and external attackers. Autonomous AI agents add another layer of risk. They can make decisions and perform actions quickly while accessing company resources.
Microsoft’s approach treats AI agents as entities that need their own identities, permissions and oversight. This can help businesses manage risk as they expand their use of AI.
However, these controls cannot eliminate every security threat. Organisations still need clear policies, regular monitoring and human supervision.
Microsoft’s official announcement explains its approach to agentic security. Businesses can also consult Microsoft’s agentic security documentation for further details.
The update highlights a growing challenge for enterprise AI. Companies must make AI agents useful while keeping their actions within clear security boundaries.
Read more : ByteDance Steps Up AI Safety Efforts
