Booz Allen Demonstrates AI Agents Rapidly Compromise OT Networks in Lab

In a Booz Allen testbed, a set of AI agents launched from a compromised perimeter host and reached an isolated industrial control network in 16 minutes. The agents completed eight distinct operational scenarios, including reconfiguring PLCs, moving physical equipment, and adapting to obstacles without human intervention.

What Happened

The lab environment mirrored a typical OT topology: a perimeter firewall, a corporate network, and a separate control network with PLCs, SCADA servers, and motion‑control devices. The AI agents began on the perimeter, enumerated exposed services, and leveraged discovered credentials to pivot into the control network. Within minutes they discovered and reprogrammed PLC logic, then physically moved a robotic arm to a new location.

Impact

While the test was isolated, the speed and breadth of the attack demonstrate that a real adversary could gain full control of an OT system in less than a quarter‑hour. The ability to reconfigure equipment on the fly could lead to production line shutdowns, safety incidents, or sabotage of critical infrastructure.

What to Do

• Segment OT traffic with strict firewall rules and enforce the principle of least privilege for all devices.
• Deploy continuous network monitoring that flags anomalous lateral movement and configuration changes in PLCs.
• Implement rapid response playbooks that include automated rollback of PLC logic and isolation of compromised hosts.
• Regularly test defenses against AI‑driven adversaries using red‑team exercises that simulate rapid pivoting and equipment manipulation.

Lessons Learned

AI agents can adapt to obstacles and complete complex OT tasks in minutes, proving that traditional perimeter hardening alone is insufficient.