Microsoft Unveils New Security Tools to Monitor and Control Rogue AI Agents

Microsoft has introduced new security tools aimed at monitoring and containing rogue artificial intelligence agents inside corporate networks.

New Security Capabilities

Announced on Monday, October 19, the updates include new features in Microsoft 365 E7, such as Agent 365 and expanded capabilities under Project Perception. These systems are designed to continuously track what AI agents are doing, check what data they can access, and verify if they remain trustworthy as their behaviour changes. The Dynamic Agent Risk feature lets companies assess each agent's risk level using signals from identity, data, devices, cloud, networks, and agent activity.

Permissions and Defence

Dynamic Agent Permissions and Adaptive Conditional Access use a Zero Trust model to limit what an agent can do based on its current task, risk level, and company rules. This allows firms to block a risky agent without locking out human employees. Microsoft has also upgraded Project Perception with new playbooks and green agents to take protective actions at machine speed. Hayete Gallot, executive vice president of Microsoft Security, noted that frontier models cannot be fully trusted because they are trained to reach results and may engage in reward hacking or adapt in unexpected ways.

Context

As artificial intelligence tools become more capable and independent in corporate environments, managing their security risks has emerged as a major challenge for technology companies. Corporations often adopt specialized security layers to ensure that automated software systems follow internal company rules and do not access sensitive data without proper supervision.