A new report from Booz Allen demonstrates that “frontier super intelligence navigated unfamiliar industrial systems, adapted when blocked, and moved physical equipment within minutes.”
The findings of the report were presented last week at SecurityWeek’s Industrial Control Systems (ICS) Cybersecurity Conference in Nashville, ahead of the public release.
Use of the term ‘super intelligence’ is, however, confusing. This is not the first time the firm has used it, but it seems to be indistinguishable from the more common term of artificial intelligence (AI). It also quickly follows Donald Trump’s September 29, 2026 Executive Order to “replace the term ‘Artificial Intelligence’ (‘AI’) with ‘Super Intelligence’ (‘SI’) in the executive branch.”
The problem, however, is that ‘superintelligence’ (one word) has long been used for a specific quality of AI. It was defined by Oxford philosopher Nick Bostrom in 2014 as, ‘Any intellect that greatly exceeds the cognitive performance of humans in virtually all domains of interest.’ It is consequently more akin or superior to artificial general intelligence (AGI), which is in turn superior to common or garden AI.
So, is Booz Allen referring to frontier AI in line with Trump’s executive order, or something more advanced in line with Bostrom’s ‘superintelligence’? The world of AI is suffering from a lack of consistent and formal definitions; but we’ll assume that Booz Allen is using ‘super intelligence’ to indicate the improving capabilities of contemporary artificial intelligence rather than a new type of AI. (SecurityWeek reached out to Booz Allen for clarification on the terminology but had received no response at the time of writing.)
The firm examined the potential effect of AI agents on OT using AI frontier models within its own isolated OT lab. These agents were given no prior knowledge of the OT in question: neither source code, nor engineering documents, nor advanced OT guidance. While the firm’s OT environment isn’t necessarily identical to enterprise OT, it is a reasonable representation.
“Across eight scenarios, every defined test objective was achieved. More important, the models connected work that usually spans multiple people and stages, including asset discovery, device research, attack planning,” summarize the researchers. More specifically, it took little more than 16 minutes for models in one test to move from a perimeter compromise to actions inside an industrial control network. In another test, a model found a robotic arm, mapped its protection zones and motion limits, and moved it within minutes.