Kill Chains and Coffee Episode 4: World’s Largest Controlled Hyperattack

In the latest episode of the “Kill Chains and Coffee” series, Armadin Offensive Security Manager Christian Elston joined me to talk about two specific kill chains, along with a wide range of topics related to AI Hyperattacks.

We discussed the power of an AI agent collective attacking an organization from the outside in, especially when those AI agents are backed by elite methodology and expert red team training. 

We also covered why safety harnesses are so critical when using AI for offensive security in the wake of the HuggingFace attack and similar incidents.

Throughout the episode, Christian shared his insights on outside-in attacks, the tradecraft behind the Armadin Attacker, and how we create safety in depth to conduct real attacks in production environments.

Kill Chain 1: SSRF > Kubernetes > Cloud Compromise

Our first kill chain came from Armadin’s initial controlled AI Hyperattack engagement, for a Fortune 600 company that invests heavily in security. We launched 100 simultaneous attacks with thousands of agents against the company’s perimeter. This amounted to nearly 1 million autonomous actions.

We discovered 12 endpoints that were vulnerable to unauthenticated Remote Code Execution (RCE) from the internet through Server-Side Request Forgery (SSRF). A combination of SRRF and a known CVE allowed our AI agents to access an overly permissive Kubernetes service.

That led the agents to escape the Kubernetes pod by identifying and exfiltrating credentials for third-party services. The agents then authenticated directly to create a kill chain that compromised the organization’s entire cloud infrastructure.

Kill Chain 2: The World’s Largest Controlled Cyberattack

Our second kill chain came from a recent engagement for a company running globally critical infrastructure, where we conducted the largest controlled cyberattack in history.

The sheer scale and magnitude of this AI Hyperattack were far beyond anything we’ve done previously. It targeted over 25,000 services across 1,300 attacks and executed millions of autonomous actions. The resulting report included over 200 findings with nearly 40 validated kill chains, representing exposure of sensitive information and the company’s crown jewels.

Key Lessons for Every Security Team

The good news is that engagements like these help you understand the blast radius around your exploitable risk. It’s just another example of how the Armadin Attacker reveals the true security posture of an environment at both speed and scale.

The more sobering takeaway is that AI Hyperattacks are scarily effective at breaching the perimeter of some of the world’s largest and most secure organizations. And we’re only getting started on how powerful these attacks can be.

Key lessons:

• AI attacks are incredibly effective at breaching the perimeter
• Armadin is running real controlled AI attacks in real environments today
• Armadin clients engage with us because they trust us to attack them safely 
  

Dive Deeper into a Controlled AI Hyperattack

Discover how a controlled assessment gives you the tools and confidence to identify truly exploitable risk across your environment. Learn more about AI Hyperattacks.