All incidents

This is a lab experiment, not a real attack. Booz Allen tested frontier AI models against an isolated industrial testbed at its lab in Chantilly, Virginia, built to resemble a manufacturing plant with real controllers, a SCADA platform, a motor drive, sensors and a robotic arm. The models were given only network boundaries, with no source code, engineering documents or OT guidance, and the company says they reached every objective across eight scenarios. In one run a model went from a perimeter breach to acting inside the control network in just over 16 minutes. In another it located the robotic arm and moved it. Existing defenses slowed or blocked some actions but no single safeguard consistently stopped the models, and the models were not named. Reporting says the AI mostly used the devices' own protocols and default passwords, not new exploits, and that the tests paused for human approval before physical actions. Dark Reading adds that Booz Allen left some unpatched vulnerabilities in place, so the timing excludes time to develop new exploits, and outlets differ slightly on what the final physical step was. The results come from the company's own report and have not been independently reproduced.