Cloud-Native AI Security Architecture for the U.S. Electric Grid

Abstract

The U.S. energy sector plays a central role in national resilience because every other critical infrastructure depends on it. 
When the grid is disrupted, the effects quickly spread into public safety and the economy. As utilities move toward cloud-based 
operations, expand telemetry, and integrate more distributed energy resources (DERs), the line between operational technology 
(OT) and IT becomes thinner. This increases both the attack surface and the risk of cyber physical infrastructure failures. 
Artificial intelligence (AI) can strengthen grid awareness by detecting anomalies, threat, forecasting issues, and supporting 
predictive maintenance. At the same time, AI brings new risks such as poisoned telemetry, adversarial inputs, model tampering, 
and supply chain vulnerabilities that can reduce trust in systems that must operate safely. 
This paper introduces a practical, Azure-aligned reference architecture for securing cloud-native AI systems in the U.S. electric 
grid. The design includes OT-aware Zero Trust connectivity, layered security controls across data ingestion, storage, training, and 
inference, and resilient edge-cloud deployment patterns that maintain reliability even when connectivity is limited. Governance is 
guided by the National Institute of Standards and Technology (NIST) AI Risk Management Framework (AI RMF 1.0), using its 
Govern, Map, Measure, and Manage functions to improve traceability, monitoring, and risk handling throughout the AI lifecycle. 
A transmission-grid anomaly detection case study demonstrates how these principles apply in real deployments, including secure 
telemetry ingestion, model registry protections, and fail safe behaviors that align with operational needs. 
The architecture also includes modern practices like tracking where models come from, AI red teaming, privacy preserving 
monitoring, and safeguards for foundation models to manage new risks from autonomous DER behavior and changing federal AI 
safety guidelines.