AI models met all eight OT attack targets in Booz Allen testbed

Booz Allen tested frontier AI models against an isolated industrial control lab in Chantilly, Virginia, and the models achieved all eight predefined attack objectives. They mapped networks, identified equipment weaknesses, wrote exploits, moved into production systems, and in one case took administrative control of a robotic arm. In a separate run, a model reached the control zone from the perimeter in just over 16 minutes, though physical actions required human approval under the test rules.