What Frontier AI Models Can Actually Do in an xOT Environment
Organizations are adopting AI faster than OT-specific guidance is being written, and adversaries are already using commercial AI against OT. To understand what that changes, OpenAI engaged Dragos to test its frontier models against real industrial targets in a controlled lab.
This technical whitepaper details what GPT-5.5-Cyber and GPT-5.6 Sol did across seven assessments on Dragos cyber ranges, industrial firmware, and ICS malware, from mapping an OT network with almost no context to finding zero-days, and where expert review still mattered.
Inside the whitepaper
How the models reached Stage 2 of the ICS Cyber Kill Chain
Zero-days found in power system engineering software
Firmware analysis and a partial PIPEDREAM module rebuild
Where the models crashed devices and produced false positives
How the SANS Five Critical Controls map to each finding
“The skill floor is dropping for the reconnaissance and vulnerability research needed before an attack.” Rob T. Lee, Chief AI Officer, Dragos
Expand your intel and cybersecurity skill sets with resources and training from your ally. Our unprecedented ICS/OT knowledge and expertise are at the ready through our robust catalogue of whitepapers, playbooks, webinars, report, and other resources for every challenge.
AI in OT: What It Means for Asset Owners and Operators
Join Dragos Chief AI Officer Rob T. Lee and CEO Robert M. Lee for a live discussion on adopting and securing AI in OT, and defending against adversaries who are already using it.
.png)
