# Microsoft Deploys AI Guardian Agents to Monitor Corporate AI Threats > Microsoft launches Guardian Agents to track corporate AI risks and prevent insider threats. - Title: Microsoft Deploys AI Guardian Agents to Monitor Corporate AI… - Summary: Microsoft launches Guardian Agents to track corporate AI risks and prevent insider threats. Microsoft introduces Guardian Agents to monitor AI agents for… - Keywords: microsoft, project perception, guardian agents, ai agents, insider threats, microsoft security, agent governance, enterprise ai, finance, markets, Deploys, Guardian - Source: BigGo Finance — https://finance.biggo.com/news/85dd8ec1-ab5a-4052-a3b7-6a6deaaf1fb5 - Author: BigGo Finance - Published: 2026-10-20T03:35:18+00:00 - Read time: 1 min - Topics: microsoft, project perception, guardian agents, ai agents, insider threats, microsoft security, agent governance, enterprise ai, finance, markets ## Microsoft expands Project Perception Microsoft is rolling out Guardian Agents to monitor corporate AI agents around the clock and contain threats when they act outside human-set policies. > Microsoft is treating the AI agents running inside corporate networks as a new class of insider threat ## Guardian Agents monitor AI activity Guardian Agents continuously track other agents, flag policy violations and recommend or execute containment steps within human-set rules. > The tool continuously monitors what other agents do, flags actions that fall outside policies set by humans, and can recommend or carry out containment steps ## Guardian Agents share signals with Microsoft Agent 365 Guardian Agents integrate with Microsoft Agent 365 to provide a unified view of agent activity for IT and security teams. > Guardian Agents shares signals, context and a unified registry with Microsoft Agent 365 ## AI agents pose new insider risks Corporate AI agents now hold system access, participate in approvals and can cause damage from inside the organization. > The core argument is that the danger is no longer only external ## Microsoft security executive warns about AI unpredictability Microsoft Security executive Hayete Gallot states frontier models cannot be trusted due to unpredictability rather than malice. > Frontier models, regardless of their origin, cannot be trusted ## Microsoft adds 35 more agent types to detection Microsoft now detects 60 local agent types including Claude Code, Codex and GitHub Copilot, up from 25. > On discovery, Microsoft is adding 35 more locally running agents to its detection list, bringing the total to 60 ## Microsoft introduces Dynamic Permissions New Dynamic Permissions model starts agents with zero rights and grants minimum access for limited time. > Dynamic Permissions starts agents with zero rights; grants minimum access for a limited time ## Microsoft's stance on AI trust Microsoft's CEO Satya Nadella argues the most trustworthy agent system requires minimal trust in the model itself. > The most trustworthy agent system is not the one built on the most trusted model ## Key takeaway Microsoft introduces Guardian Agents to monitor AI agents for policy violations and contain threats.