Powered by MOMENTUMMEDIA
For breaking news and daily updates, subscribe to our newsletter

CrowdStrike points AI agents at enterprise weaknesses with autonomous red teaming

The company has introduced an AI-powered offensive security service that emulates agentic adversaries, with early access available to select customers.

• Thu, 05 Nov 2026 •
CrowdStrike points AI agents at enterprise weaknesses with autonomous red teaming

CrowdStrike is using purpose-built AI models to automate red-team exercises, identifying how attackers could combine vulnerabilities and security weaknesses to reach an organisation’s critical assets.

The new Agentic Adversary Emulation service, powered by the company's SafeMind security models, operates like an autonomous attacker in controlled, customer-approved environments.

CrowdStrike said the approach is intended to address the limitations of point-in-time penetration tests and periodic security assessments, which can struggle to keep pace with adversaries using AI agents to discover and exploit weaknesses at machine speed.

 
 

“The agentic adversary uses AI to operate at a speed and scale no human can match. These are preventable gaps that go unprevented,” Adam Meyers {pictured], head of counter adversary operations at CrowdStrike, said in a statement.

“Now defenders can find and close them before the adversary does. That is offence in the service of defence.”

Agentic Adversary Emulation uses Red Tempest, SafeMind's purpose-built offensive AI model, which is designed to operate like an autonomous attacker. Its supporting harness can orchestrate thousands of steps, chaining weaknesses, adapting around defences and pursuing defined objectives during authorised testing.

The service combines that capability with CrowdStrike intelligence covering more than 290 tracked adversaries, emerging agentic threats, and newly disclosed vulnerabilities. It also draws on Falcon environmental context, including information about vulnerabilities, misconfigurations, and identity gaps.

SafeMind itself is trained using Falcon sensor telemetry, expert-labelled threat intelligence and operational experience from CrowdStrike analysts and incident responders.

The platform can generate recommended compensating controls, including patches, configuration changes and new detections, before Red Tempest re-runs the simulated attack to validate whether the path has been closed.

Agentic Adversary Emulation is currently available in early access to select customers through Project QuiltWorks partners.

Cyber DailyWant to see more stories from trusted news sources?
Make Cyber Daily a preferred news source on Google.
Tags: