Microsoft unveils controls to monitor and limit enterprise AI agents

Microsoft announced a set of security controls for enterprise AI agents, designed to monitor their activity and reduce or stop their access when their assessed risk rises. The company said the capabilities will be included in Microsoft 365 E7 and its Project Perception security system; it did not provide availability dates or pricing in the announcements covered here. The controls are intended to address the possibility that an agent’s actions may shift as its instructions, data or working context change. Microsoft says Dynamic Agent Risk will continuously assess agents using signals from identity, data, devices, cloud, network and agent activity. Under Dynamic Permissions, an agent would receive access for a specific task and only for the time needed to complete it. Dynamic Conditional Access would adjust that access as risk changes. Depending on the situation, an agent could lose permission to make changes while retaining read access, be restricted from certain data or systems, require human approval for an action, or be stopped. Microsoft also said it has expanded discovery to cover 35 additional types of agents running on employee devices, bringing the total to 60....