Policy
DOE deploys AI to map grid vulnerabilities, secure bases
New foundation models analyze billions of contingencies while Project Armor identifies weak points in defense installations.
Key takeaways
- The Energy Department's CESER is using AI to shift from defensive to offensive grid security, identifying vulnerabilities in programmable logic controllers and other embedded systems.
- Project Armor maps energy vulnerabilities at critical military installations as part of DOD's Fortress initiative to ensure defense energy independence.
- Argonne National Laboratory's GridAI foundation model performs contingency calculations 4,000 times faster than conventional solvers, covering 99% of grid risk scenarios.
- AI-assisted assessments now deliver real-time situational awareness to CESER, replacing processes that previously took up to a year.
- GridAI is trained on data from three national laboratory supercomputers and models power system changes on a second-by-second basis.
The U.S. Department of Energy is deploying artificial intelligence systems to identify vulnerabilities in the power grid and harden critical military infrastructure against both natural disasters and adversarial attacks.
At AFCEA Bethesda's Energy, Infrastructure and Environment Summit, DOE officials detailed how AI is reshaping their approach to energy security—moving from reactive defense to proactive threat modeling across interconnected systems.
Why it matters
The electric grid underpins every aspect of military readiness and national security. As adversaries develop more sophisticated capabilities to target critical infrastructure, the ability to model billions of failure scenarios in real time—rather than waiting months for manual assessments—could mean the difference between resilience and catastrophic cascading outages during a crisis.
From defensive to offensive security posture
Andrew McClure, director of the Office of Cybersecurity, Energy Security and Emergency Response (CESER), explained that many grid components are inherently insecure by design. Programmable logic controllers, battery management systems, and telecommunications devices embedded in power transformers create multiple attack surfaces.
The challenge, McClure said, is ensuring AI inference models don't inadvertently increase risk. "After we train models, we're going to have increased insecurity across the grid," he noted. "We need to invest in the foundational research as well as the mature technologies that are commercially ready."
Project Armor targets military installations
CESER has launched Project Armor, a partnership with the Defense Department's Energy, Installations and Environment Office, to map vulnerabilities at critical military sites. The initiative supports DOD's broader Fortress program, which aims to ensure energy independence for defense installations.
McClure said the work involves establishing energy supply baselines for different bases and answering fundamental questions: Does an installation need additional bulk fuel storage? Are substations feeding the base resilient enough? How should grid dependencies be modeled?
"Those are the kinds of scenarios that we think about and provide a boutique set of solutions for these military installations," McClure said.
GridAI analyzes billions of contingencies
Researchers at Argonne National Laboratory have developed GridAI, the first foundation model designed to understand the interconnected systems powering the grid. Trained on data from three national laboratory supercomputers, the platform accounts for power system changes on a second-by-second basis.
Duane Rudolph Verner, manager of Argonne's Global Energy Security Department, said GridAI performs computations 4,000 times faster than conventional utility solvers and covers 200 times the outage risk aperture. The model can evaluate tens of billions of contingency calculations, covering 99% of the overall risk landscape.
Real-time assessments replace year-long reports
AI is also accelerating how national laboratories provide situational awareness to CESER. Verner noted that traditional security assessments required field teams and could take a year to produce final reports. AI-assisted tools now enable faster assessments and reporting, delivering actionable intelligence to decision-makers when it matters most.
These details were first reported by Federal News Network.