Discover AI agents and assess security posture using Microsoft Defender

Microsoft Defender provides a centralized inventory of AI agents in your organization and assesses their security posture.

The inventory includes:

• Agents built with Microsoft Copilot Studio, Microsoft Foundry, Microsoft 365, and supported non-Microsoft platforms.
• Local AI agents discovered on endpoint devices.

Use the AI Agents page in the Microsoft Defender portal to review agent configuration, risk levels, risk indicators, recommendations, alerts, tools, identities, and related security context. You can also query agent inventory and configuration data by using Advanced Hunting.

Prerequisites

• Enable security for AI agents, including the Microsoft 365 connector. See Enable security for AI agents using Microsoft Defender.
• To discover local AI agents that run on endpoints, set up AI agent runtime protection in Microsoft Defender for Endpoint. Discovery requires Microsoft Defender for Endpoint and Microsoft Defender Antivirus in active mode. Local agents are onboarded separately.

Review AI agent risk and security posture

Microsoft Defender calculates an agent's risk level from its active risk indicators. Security recommendations identify available actions for improving the agent's posture but are calculated separately from the risk level.

The available risk indicators, recommendations, tabs, and supporting evidence depend on the agent type and the security information available for that agent.

To review the risk and security posture of AI agents:

1. Sign in to the Microsoft Defender portal.

2. In the left navigation pane, select Assets > AI agents.

3. Select the Agents tab.

   The inventory includes the following risk-related columns:

   • Risk level: The overall risk level calculated from the agent's active risk indicators.
   • Risk indicators: The conditions contributing to the agent's risk level.
   • Recommendations: The number and severity of active security recommendations for the agent.
   • Active alerts: The number and severity of active alerts associated with the agent.

   

4. Use the filter bar to narrow the inventory by properties such as Agent name, Platform, Publish status, Risk level, or Risk indicators.

5. Select Customize columns to add, remove, or reorder columns in the inventory.

6. Select an agent to open its details pane.

   The pane displays the risk and configuration information available for the selected agent.

   

7. Select Open Agent page to review more information about the agent.

   The Overview tab shows the security and configuration information available for the agent, which can include:

   • Risk level and active risk indicators
   • Agent configuration details
   • Identity and authentication information
   • Endpoint and user context, when available
   • Attack-surface relationships
   • Active alerts
   • Security recommendations
   • Tools
   • MCP servers

8. Review any additional tabs available for the agent:

   • Select Security recommendations to review posture recommendations associated with the agent. Select a recommendation to review its description and supporting evidence on the Overview tab, corrective actions on the Remediation steps tab, and affected agents on the Exposed assets tab.
   • Select Incidents and alerts to review security incidents and alerts associated with the agent. Select an incident to open its details pane, or select Open incident page for the full investigation experience.

   The available tabs depend on the agent platform and the security information associated with the agent.

For details about risk indicators, risk levels, and recommendation logic, see AI agent posture risk in Microsoft Defender.

For more information about investigating agent threats, see Detect and investigate threats to AI agents using Microsoft Defender.

Discover and assess AI agents using Advanced Hunting

The AgentsInfo table in Advanced Hunting provides an inventory of AI agents and their security-relevant properties.

Use the table to:

• Discover AI agents in your environment.
• Review available agent properties, such as authentication, access control, tools, knowledge sources, orchestration settings, endpoint context, and user context.
• Correlate agent inventory information with other security data.

Note

The AgentsInfo table replaces the previous AIAgentsInfo table as part of the Microsoft Agent 365 transition. For more information, see Transition Microsoft Copilot Studio and Microsoft Foundry agent security capabilities to Microsoft Agent 365.

To query agent inventory data:

1. Sign in to the Microsoft Defender portal.

2. Select Investigation & response > Hunting > Advanced hunting.

3. Query the AgentsInfo table.

   To use the prebuilt queries maintained by Microsoft, select the Queries tab, and then select AI Agents. For more information, see Sample queries.

   To return the latest record for each agent, run the following query:

   AgentsInfo
   | summarize arg_max(Timestamp, *) by AgentId
   | where LifecycleStatus != "Deleted"
   

Important

The AgentsInfo table stores multiple snapshots of each agent over time. Use arg_max(Timestamp, *) to return the latest state of each agent. For more information, see arg_max() aggregation function.

For more information:

• To query local agents discovered on endpoint devices, see Discover local AI agents in Microsoft Defender for Endpoint.
• For information about Advanced Hunting, see Proactively hunt for threats with Advanced Hunting in Microsoft Defender.

Next steps

• AI agent posture risk in Microsoft Defender
• Detect and investigate threats to AI agents using Microsoft Defender
• Protect AI agents in real time using Microsoft Defender
• Protect AI assets from emerging threats and vulnerabilities using Microsoft Defender
• Microsoft Defender for Cloud AI security posture management