Microsoft and Sumsub treat AI agents as insider threats, forcing zero-trust controls

On the same day, Microsoft and Sumsub announced separate security products that converge on treating AI agents as insider threats rather than trusted software. Microsoft introduced Guardian Agents under its Project Perception system, which continuously monitor every action of deployed agents before it happens, along with Dynamic Permissions that grant agents only task-specific access and Dynamic Agent Risk that rates agents using identity, data, device, cloud, network, and behavioral signals. Microsoft Security EVP Hayete Gallot said agents can reach sensitive systems, are often over-provisioned, and can behave unpredictably or even collude with other agents, so frontier models cannot be trusted. Sumsub's new KYA Module for Model Context Protocol screens every agent request at the data-access layer, stripping personally identifiable and sensitive information before it reaches the agent, and can trigger real-time liveness checks to confirm a human is behind high-risk actions. This follows Sumsub's earlier AI Agent Verification and MCP integration, and research it cited found 65% of organizations experienced at least one security incident involving an AI agent in the past year. The convergence marks a shift from treating agents as tools to treating them as personnel, with both vendors racing to define a new agent-governance control plane that enterprises will need as deployments accelerate.

(Source：Vector Wire)