Autonomous AI systems successfully executed cyberattacks against critical infrastructure in all eight test scenarios, according to a report published by Booz Allen Hamilton on October 11, 2026. The advanced models turned digital access into physical disruption — locating and moving a robotic arm in just minutes in one case, and advancing from a perimeter breach to actions inside an industrial control network in just over 16 minutes in another. The consulting firm's operational technology lab tested whether the latest frontier AI models could identify, infiltrate, and manipulate industrial equipment that controls water, power, and other essential services.
The tests placed two of the latest frontier models from leading AI providers into a multi-vendor environment modeled after a general manufacturing facility, complete with programmable logic controllers, human-machine interfaces, SCADA platforms, variable-frequency drives, robotic arms, and other physical equipment. The models received no source code, engineering documents, or advanced technical guidance, forcing them to conduct their own research, attack planning, and execution. Testers built guardrails requiring human approval before exploiting security issues or taking actions with physical impact. The lab environment used mixed vendors, firmware, control logic, and imperfect segmentation to reproduce the complexity common in long-lived operational technology settings. The eight tests examined whether the models could map environments, find vulnerabilities, exploit weaknesses, move laterally into production systems, manipulate controllers, alter motor speeds, compromise SCADA systems, and control robotic equipment.
In the SCADA test, agents initially targeted the wrong version of the operator interface, then adjusted their strategy by checking active sessions and determining that the control room used a different HMI client version. The model found editable code in the exported SCADA project, rebuilt its payload, and used the administrative interface to distribute a full-screen takeover of the interface visible to operators in the control room. According to Kyle Miller, vice president of infrastructure cybersecurity at Booz Allen, "Our testing showed that AI agents can operate with a speed, persistence, and engineering-level precision that may outpace organizations that have not implemented foundational OT cybersecurity practices." The report states that the models performed operational technology tasks "with a high degree of engineering-level precision," repeatedly producing intended controller and equipment actions across multiple vendors. In the robotic arm test, an agent probed the network for common robotic protocols, identified the robot, discovered its API, gained administrative access, mapped protection zones and motion limits, and moved the arm — all within minutes.
The research demonstrates that criminals no longer need specialized operational technology expertise to exploit weaknesses in industrial systems, potentially allowing less-skilled attackers to cause physical disruption. OT systems have historically relied on obscure protocols and proprietary hardware and software for protection, but agents can now simply learn everything about these systems on demand. Many of these devices don't use authentication or encryption once someone breaks in, making it easier for autonomous agents to execute malicious commands. "Specialized OT knowledge, unfamiliar equipment, and complex control environments are no longer meaningful barriers to attack," the report concludes. The consulting firm is calling for more industry testing, development, and increased deployment of cyber defenses across operational technology and other critical infrastructure networks. Miller noted that OT security maturity varies significantly across industries, with many critical infrastructure organizations continuing to face challenges implementing security controls across siloed, highly variable, and globally distributed environments. The defensive window may be shrinking fast — organizations could have only minutes to detect and block an autonomous attack once it begins. The shift from theoretical vulnerability to demonstrated capability means the infrastructure that keeps daily life running now faces adversaries that operate faster than human defenders can respond, and the gap between attack speed and organizational readiness may determine which systems survive the next decade intact.