Detect Threats to AI Agents

Turns agent activity into detections — surfacing prompt injection, jailbreak, credential theft, and abnormal execution against agents and the AI infrastructure they depend on, with alerts centralized for the SOC.

Core capabilities

Agent Threat Detection Microsoft Defender XDR

Detect threats targeting agents — abnormal execution, risky tool use, and suspicious activity — and raise alerts in the Microsoft Defender portal
AI Threat Protection Microsoft Defender for Cloud

Detect runtime attacks on the underlying Azure OpenAI and Azure AI model inference workloads — prompt injection, jailbreak, data leakage, and wallet abuse — and centralize alerts in Defender XDR
Agent Runtime Protection Status Microsoft Copilot Studio

Confirm that Copilot Studio agents are covered by runtime protection so detections aren't blind to low-code agents

Supporting capabilities 2

Purview DSPM AI Interaction Discovery Microsoft Purview Data Security Posture Management

Flag sensitive data flowing into agent prompts so detections can be prioritized by data-exposure risk
Agent Security Posture Management Microsoft Defender XDR

Reduce the attack surface ahead of detection by identifying misconfigurations and posture gaps across agents

Core capabilities

Solutions

Further reading