Booz Allen Report: AI Agents Control Unfamiliar OT Systems and Move Equipment in Minutes

AI Agents Demonstrate OT Control

A new report from Booz Allen, presented at SecurityWeek’s Industrial Control Systems (ICS) Cybersecurity Conference, details the ability of AI agents to interact with operational technology (OT) systems. The report states that "frontier super intelligence navigated unfamiliar industrial systems, adapted when blocked, and moved physical equipment within minutes."

Methodology and Findings

Booz Allen examined the potential effect of AI agents on OT using AI frontier models within its isolated OT lab. These agents were given no prior knowledge of the OT system, including source code, engineering documents, or advanced OT guidance. The firm's OT environment is considered a reasonable representation of enterprise OT.

Across eight scenarios, all defined test objectives were achieved. The models connected work that typically involves multiple people and stages, such as asset discovery, device research, and attack planning. In one test, models moved from a perimeter compromise to actions inside an industrial control network in just over 16 minutes.

Terminology Clarification

The report uses the term 'super intelligence,' which has caused some confusion. While the term 'superintelligence' (one word) was defined by Oxford philosopher Nick Bostrom in 2014 as an intellect greatly exceeding human cognitive performance, Booz Allen appears to use 'super intelligence' to indicate the improving capabilities of contemporary artificial intelligence rather than a new type of AI. This usage aligns with a recent executive order to replace 'Artificial Intelligence' with 'Super Intelligence' in the executive branch.

Implications for Industrial Control Systems

The findings highlight the growing sophistication of AI in interacting with and potentially controlling critical infrastructure. The ability of AI agents to quickly adapt and execute complex tasks in unfamiliar OT environments suggests both opportunities for automation and potential security risks if such capabilities were to be misused.