The Pulse
CrowdStrike Adds Agent-Insider Detection to Its Agentic SOC
CrowdStrike says its Falcon platform can connect an AI agent’s activity to its owner and device, then investigate when those signals point to compromise. The company also describes approval controls and response tools for managing agents ac
AI.info Team · Written with AI · fact-checked ·
CrowdStrike’s Nov. 4 announcement puts two competing realities side by side: an AI agent may be an approved part of an employee’s work, yet become an insider threat if its behavior changes. The company says its Falcon platform can connect an agent’s activity to its owner and device, then investigate and respond when those signals point to compromise.
The risk, CrowdStrike writes, is an agent running under an employee’s identity with approved tools and access, but drifting after a change such as a tampered tool or an unexpected instruction. The company argues that activity may look routine in isolated alerts, while a pattern linked to the same agent, person and machine tells a different story.
Falcon connects an agent to its owner
CrowdStrike criticizes two existing approaches: agentic SOC tools that run a single language model through detections without prior investigation or identity context, and standalone AI security products that flag unusual activity but stop at detection. Its proposed system combines agent monitoring with specialist agents that investigate and respond across the Falcon platform.
Agent Insider Threat Detection, listed for Q4, uses Falcon Next-Gen SIEM to learn normal behavior for each AI agent and its operator, including the files the agent touches, the tools it calls and where it sends data. The product is designed to flag deviations even when an agent has not violated its granted permissions.
Dynamic Risk Scoring, also listed for Q4, feeds agent risk into the same 0-to-100 score Falcon Next-Gen SIEM builds for the agent’s owner and host. CrowdStrike says seeing the agent, user and device rise together on the risk list can make coordinated activity easier to spot.
AI Ecosystem Detections, listed for Q4, cover attacks on AI services, including LLM API abuse, stolen API keys and compromises of coding assistants, MCP servers and agent frameworks. Federated Search Expansion adds connectors for GovCloud, Dynatrace, Databricks and Snowflake, which CrowdStrike says are generally available; connectors for Splunk, Elastic and Cribl Lake are due in Q4.
Investigations trace what the employee asked for
Multi-Agent Triage and Investigation is generally available, with CrowdStrike saying it was announced at Fal.Con US 2026. Specialist SOC agents investigate detections together using shared context across endpoint, identity and agent activity.
In Q4, the company plans to extend triage coverage to Falcon Guardian, Falcon Cloud Security, Falcon Shield, Falcon Data Security and custom detections, including abnormal agent behavior flagged by Falcon Next-Gen SIEM. When a detection reaches the investigation, CrowdStrike says the agent, its owner and its host are already connected.
The investigation can trace an agent session back to the person’s request and distinguish it from actions the agent took independently. CrowdStrike’s stated aim is to determine whether the employee directed the activity or whether the agent acted without the employee’s knowledge.
The Unified Agentic SOAR Experience is generally available and was also announced at Fal.Con US 2026. Teams can build and govern automation in plain language, set triggers and rules, and decide which actions run autonomously and which require approval.
Approval gates can stop an agent mid-task
CrowdStrike says a response can coordinate steps such as stopping an agent, cutting off its access, preventing it from returning and documenting the incident. The system connects to agents, models and tools through APIs or MCP, according to the company.
Human-in-the-Loop Action Requests, generally available, pause a Falcon-native or AgentWorks agent when it invokes a sensitive tool. An analyst can approve the action, reject it with a documented reason or inspect the agent’s trace before it runs.
Approvals for Agents and Workflows, listed for Q4, extend sign-off requirements to an entire agent, app or workflow. Agent Interruption, generally available, lets analysts pause or stop a running SOC agent mid-task.
Policy-Drift Agent Guardrails, powered by Falcon Guardian and listed for Q4, are designed to screen CrowdStrike Charlotte AI and AgentWorks interactions for policy drift, unauthorized tool calls and out-of-bounds data access. AI Credit Visibility and Control, also listed for Q4, gives administrators real-time visibility into agent and workflow credit use, the ability to halt consumption mid-run, and approval over raising a limit when an agent reaches its cap.