CrowdStrike has announced two additions to its Falcon platform aimed at security risks created by autonomous AI agents: “Agent Insider Threat Detection” within its agentic SOC offering, and a new “Agentic Adversary Emulation” capability designed to simulate attacker behaviour to identify exploitable paths.
The announcements were made at Fal.Con Europe 2026 in Barcelona, alongside an event dateline listing Austin, Texas, and are positioned around the idea that AI agents can create insider-style risk when they are manipulated, misconfigured or drift from intended behaviour.
According to CrowdStrike, Agent Insider Threat Detection is intended to identify abnormal AI agent behaviour “even when no policy is violated” by learning baseline activity for both an agent and its operator, then routing deviations into “coordinated, multi-agent investigations” across endpoint, identity, SaaS, cloud and network domains.
“AI agents can be both an insider threat and an adversary,” said Michael Sentonas, president of CrowdStrike. “One acts without authorisation from inside your walls. The other is weaponised against you from outside. Stopping the insider means knowing the moment an agent turns, investigating across every domain, and shutting it down at machine speed. That’s the next evolution of the agentic SOC.”
CrowdStrike said coordinated multi-agent investigations are now generally available, and listed additional controls it says are available today, including a “single-governed automation workspace” through Charlotte Agentic SOAR, “human-in-the-loop” approval gates for sensitive tool use, and the ability for analysts to pause or stop a running agent mid-task.
The company also outlined a roadmap for further features, including Agent Insider Threat Detection and “AI ecosystem detections” in Q4, as well as “policy-drift agent guardrails” in FY28 Q1. It said its “Falcon Guardian AI Kill Switch” can terminate an AI agent, and described planned capabilities such as real-time AI credit visibility and expanded federated search connectors.
Separately, CrowdStrike introduced Agentic Adversary Emulation, which it said uses its SafeMind security models to automate red-team style testing by chaining vulnerabilities and weaknesses into paths that could reach critical assets, then driving remediation and re-running the simulated attack to validate the path is closed.
“The agentic adversary uses AI to operate at a speed and scale no human can match. These are preventable gaps that go unprevented,” said Adam Meyers, head of counter adversary operations at CrowdStrike. “Now defenders can find and close them before the adversary does. That is offense in the service of defence.”
CrowdStrike linked the announcement to its 2026 Agentic AI Campaign report, which it said documented an agent chaining a vulnerability into cluster-admin access and exfiltrating 726 databases “with no human at the keyboard”, arguing that periodic assessments and point-in-time penetration tests may not keep pace with machine-speed exploitation.
The company said Agentic Adversary Emulation is available in early access for select customers through Project QuiltWorks partners.
