CrowdStrike Adds AI Agent Insider Threat Detection to Falcon

A new kind of insider threat

CrowdStrike describes an agent insider threat as a trusted AI agent acting on behalf of an employee. The agent may run under a real employee's identity, connect to tools approved by the employee's team and retain the access required for its work.

According to CrowdStrike, the risk emerges when something upstream changes, such as a tampered tool or an instruction that should not be present. The agent can then drift from its expected behavior without immediately triggering an alarm.

The company says AI agents now have two roles in security operations: they are a new operating unit of the SOC and a new identity that must be secured.

Falcon links behavior to identity and context

CrowdStrike says many agentic SOC tools run a single large language model serially through detections, without retaining context from earlier investigations or combining identity, endpoint and ownership information. In the company's view, that makes it difficult to distinguish routine activity from a compromise.

Standalone AI security tools can inventory agents and identify unusual activity, CrowdStrike says, but they stop at detection and leave another tool to handle the response. The company says its Falcon platform is designed to secure agents in the environment while also putting specialist agents to work inside the SOC to investigate and respond to threats at machine speed.

The announcement introduces Agent Insider Threat Detection, listed as Q4. CrowdStrike says Falcon Next-Gen SIEM learns the normal behavior pattern for each AI agent and its operator, including the files the agent touches, the tools it calls and where it sends data. It can then flag deviations even when the agent has not violated a policy.

Risk scoring and multi-agent investigations

CrowdStrike says agent insider threats can be difficult to identify because individual signals may resemble ordinary activity by an AI coding agent. When those signals are spread across separate alerts, they can be dismissed. The company says tying them to the same agent, person and machine can reveal a different pattern.

Another capability, Dynamic Risk Scoring, is also listed as Q4. It feeds agent risk into the same 0 to 100 risk score that Falcon Next-Gen SIEM already builds for the agent's owner and host. CrowdStrike says this approach helps teams understand the combined risk picture when an agent, its user and its device rise together on the risk list.

The wider agentic SOC update includes Agentic Investigations, which brings specialist agents together on shared context to investigate across domains and converge on one verdict, according to CrowdStrike.

What the update means for security teams

The announcement positions AI agents as both a source of additional security telemetry and a part of the response process. For security teams dealing with hundreds of thousands of alerts a month, CrowdStrike's stated aim is to combine activity signals with identity and device context rather than assess each alert in isolation.

The company says the Falcon platform will let teams detect abnormal agent behavior, connect human and non-human risk, and coordinate investigations across domains. The source identifies Agent Insider Threat Detection and Dynamic Risk Scoring as Q4 capabilities, without providing further timing or availability details.