Booz Allen Hamilton’s operational technology lab ran two unnamed frontier AI models through eight scenarios testing whether they could execute an autonomous attack chain against industrial systems, and the models achieved the objectives across all eight, The Register reported Saturday. In one test, an agent probed a network, identified a robotic arm, discovered its API, gained administrative access, mapped its protection zones and motion limits, and moved the arm, all in minutes. In another, the models progressed from a perimeter compromise to actions inside an industrial control network in just over 16 minutes.
The lab replicated a multi-vendor manufacturing environment with programmable logic controllers, human-machine interfaces, SCADA platforms, variable-frequency drives and sensors. The models received no source code, engineering documents or advanced OT guidance; they were required to wait for human approval before exploiting issues or causing physical impact. The firm found the models repeatedly produced intended controller and equipment actions, altered process values and made controlled kinetic changes, and concluded that specialized OT knowledge is “no longer a meaningful barrier to attack.” Kyle Miller, Booz Allen’s vice president of infrastructure cybersecurity, said defenders may have only minutes to detect and block such attacks.
The tests follow real-world incidents in which frontier AI models gained unauthorized access to external systems, and reports of suspected Chinese operators using AI agents against South Korean financial institutions and Taiwanese government websites, and suspected Iranian actors using AI-generated scripts against exposed PLCs at US water, manufacturing and energy facilities. OpenAI, Anthropic and Google have recently announced initiatives giving critical infrastructure operators access to advanced models for defense.