Microsoft adds Guardian Agents for continuous oversight of enterprise AI agents
Microsoft expanded Project Perception to discover enterprise AI agents, constrain them with separate identities and dynamic permissions, and respond to changing risk through Guardian Agents.
AT A GLANCE
Key points
- Local-agent discovery expands by 35 types to 60 in total
- Each agent receives a separate identity and task-scoped authority
- Dynamic Agent Risk continually updates risk from multi-layer signals
- Guardian Agents act within human-defined policy and approval boundaries
What is changing?
Microsoft expanded Project Perception on October 19 to secure AI agents operating inside enterprises. The system discovers agents from different platforms, brings them into a unified registry, and creates an accountable identity for each agent that is distinct from the user. Microsoft says local-agent discovery is expanding by 35 types to 60 in total, including Claude Code, OpenAI Codex, and GitHub Copilot.
The new Dynamic Permissions approach limits an agent’s authority to the approved task and duration. Dynamic Agent Risk continuously updates a risk score using signals from identity, data, endpoints, cloud, network, and runtime. Dynamic Conditional Access then reassesses access as risk changes.
New Playbooks and a team of Guardian Agents in Project Perception can investigate evidence and recommend or apply containment within policies and approval boundaries defined by people. Microsoft says people retain control over objectives and consequential decisions.
Explore the original sources
Source date: 19 October 2026.
Published by AI Radar: 19 Oct 2026, 18:44 (Istanbul, UTC+3). Updated: 19 Oct 2026, 20:11.
FOLLOW THE TOPIC
Topics and tags
Share your discovery.
Send this story to a friend.
Reader comments
What do you think about this development? Share your experience or question. Comments are published after editorial approval.
Your name/nickname and comment become public if approved. Read: Privacy notice.
Loading comments…