AI systems are fully capable of carrying out nightmare attacks against infrastructure and nobody's ready

AI systems can now execute sophisticated attacks on critical infrastructure, significantly lowering the barrier to entry for malicious actors to cause physical disruption.

Booz Allen Hamilton's operational technology lab conducted tests on eight scenarios using advanced AI models to assess their capabilities in attacking critical infrastructure. Across all scenarios, the AI models successfully demonstrated the ability to achieve objectives, transitioning from digital access to physical actions, including controlling robotic arms and manipulating industrial control systems. In one test, an AI agent progressed from a perimeter compromise to actions within an industrial control network in just over 16 minutes. The models operated with speed, persistence, and engineering-level precision, identifying vulnerabilities, gaining access, and manipulating equipment like variable-frequency drives and robotic arms. The testers found that specialized OT knowledge and complex environments were no longer significant barriers for AI attackers, as the models could quickly learn about and exploit obscure protocols and proprietary hardware without explicit human guidance. The AI agents were able to compromise SCADA systems, alter operator screens, and control connected equipment, highlighting the potential for AI to enable less-skilled attackers to cause physical disruption.

The Booz Allen Hamilton report indicates a significant gap in readiness for AI-driven attacks against operational technology (OT) and critical infrastructure. The speed at which AI models can identify vulnerabilities, gain access, and execute physical actions—sometimes in minutes—outpaces current organizational defense capabilities, especially those not adhering to foundational OT cybersecurity practices. This suggests an urgent need for accelerated development and deployment of specialized defenses for these environments.

This capability shift means that even attackers without deep expertise in OT systems can potentially orchestrate disruptive attacks. The AI's ability to learn and adapt to proprietary systems and obscure protocols bypasses traditional security measures that relied on specialized knowledge. Organizations managing critical infrastructure must now prioritize rapid testing and implementation of AI-aware security protocols to mitigate this escalating threat.