Blog

As utilities accelerate the adoption of AI at the grid edge, the conversation is shifting rapidly from experimentation to operational reality. Edge intelligence can offer faster fault detection, improved reliability, and more efficient operations, yet it can also introduce new risks. When AI systems begin to influence grid operations in near real time, trust, security, and resilience are no longer abstract concepts; they become prerequisites for safe operation.

These themes took center stage at a recent panel discussion held during the Anterix Active Ecosystem pre-DTECH conference. Moderated by Jean Jones of Anterix, the session brought together experts from across the utility, cybersecurity, connectivity, and device ecosystems, including Mitch Rappard (Palo Alto Networks), Mitesh Parikh (GE Vernova), Jeremy Wirsche (Trident Industries), Jeff Pickles (Green Grid Inc.), and Jeremy Gosteau (Sequans). While perspectives varied, the discussion crystalized around one key reality: secure, reliable connectivity, along with data integrity and trusted networks and devices form the foundation that ultimately determines whether edge AI strengthens or undermines grid operations.

From Centralized Intelligence to Millions of Edge Data Sources

Jean Jones kicked off the panel discussion, focusing on how utilities are shifting from a handful of centralized systems to millions of intelligent edge nodes, including meters, line sensors, substation devices and systems, and other field assets, each generating data and supporting more informed operational insight. AI is accelerating this shift by enabling local analytics and advanced data filtering, while reducing latency and dependence on centralized processing.

Importantly, not all AI deployments imply autonomous decision-making. Utility AI systems may be more deterministic or assistive rather than agentic, designed to help humans prioritize, interpret, and act on vast volumes of data. Jeff Pickles framed it simply: “AI does not replace operators; it helps surface the needle in the haystack, enabling faster and more effective human decision-making.” For information to be trusted, device and network security are paramount to large-scale deployment.

Panelists highlighted practical edge AI applications already delivering value today, including computer vision for vegetation management, construction quality assurance, and safety compliance, as well as local anomaly detection on high-sample-rate sensor data such as voltage and current waveforms captured by smart meters. Jeremy Wirsche underscored that “operators can now respond to work sites fully knowing the terrain and infrastructure conditions prior to arrival, significantly enhancing both operator safety and productivity.” These workloads also benefit from processing data close to the source, reducing latency while limiting the movement of sensitive raw data.

But distributed intelligence still changes the risk profile. A compromised or poorly managed edge device is no longer just a data quality issue; it can introduce blind spots, misinformation, or delayed response across operations. As AI-enabled analytics become embedded across AMI 2.0, line monitoring, substation automation, and vegetation management, utilities must assume that every edge endpoint participating in operational workflows is part of the critical infrastructure security perimeter.

Mitch Rappard emphasized the fact that utilities leveraging AI must have security built in from day one.  He also added that a security platform approach gives utilities visibility and Zero Trust controls which ultimately lead to both better security and business outcomes.

Why Zero Trust Must Be Considered Across the Network and the Device

Zero Trust architectures only work if they extend from the network all the way down to the device itself. Network-level controls and cloud security platforms are necessary, but insufficient, without strong device identity, secure boot, and software integrity.

This holds true whether AI runs on edge hardware, in utility data centers, or in the cloud, and even in systems that do not yet use AI at all. The common denominator is secure connectivity. As Mitesh Parikh explained, utilities “operationalize Zero Trust principles through tiered network architectures, applying connectivity, compute and security controls proportional to device criticality.” In this model, Private Mobile Networks and the modems or modules that attach devices to them become the enforcement point for Zero Trust principles: identity, least-privilege access, segmentation, and continuous verification.

Panelists also emphasized that Zero Trust in utility environments must be OT-aligned (Operational Technology). Best practices discussed included keeping private LTE cores within OT domains to reduce attack surfaces, using SIM-based identity as a hardware-rooted trust anchor, enforcing granular authorization, and validating data before it is consumed by analytics or AI systems.

For utilities, this requires cryptographic identity anchored in hardware, often implemented through cellular modules, SIMs, or embedded secure elements. When every device can attest to its identity, the firmware it is running, and its known-good state, utilities gain the ability to enforce meaningful policies: determining which devices can connect, what data they can send, and what actions they are permitted to trigger.

This becomes even more critical as AI models, analytics pipelines, and operational workflows evolve. Continuous verification of identity, software, and behavior is essential as threat landscapes change over time.

Private LTE and 5G as Utility-Grade Foundations

Secure and resilient connectivity emerged as another foundational theme. AI-enabled grid operations depend on predictable, utility-grade network performance, particularly for time-sensitive or safety-critical use cases. Private Mobile Networks provide deterministic coverage, latency, integrated identity, and quality of service that public networks alone cannot guarantee.

With the added protections of a private-utility controlled cellular network, utilities gain a robust connectivity fabric that maintains visibility and control even during storms, congestion, or cyber events. This resilience becomes increasingly important as analytics-driven insights move closer to operational workflows.

At the same time, low-power technologies such as LTE-M and NB-IoT play a complementary role. By pairing efficient edge analytics with LPWA connectivity, utilities can process data locally, transmit only enriched insights, conserve power, and reduce unnecessary exposure of sensitive raw data, an approach well suited for large-scale, remote sensor deployments.

Designing for 20-Year Lifecycles

A recurring challenge is time. Utility devices often remain in service for 15 to 30 years, while AI techniques, operational requirements, and cyber threats evolve much faster. Jeremy Gosteau stressed that “without secure boot, hardware root of trust, and reliable, secure firmware-over-the-air capabilities, today’s innovation can become tomorrow’s vulnerability”.

In this context, panelists underscored emerging AI-specific risks, such as data integrity issues, model drift, particularly as analytics move closer to operational decision support and, in limited well-defined cases, toward more autonomous behavior. Erroneous or manipulated models can introduce cyber-physical risk, where digital failures translate into real-world grid impacts. Defending against these threats increasingly requires automated, AI-enabled security tools, as human-only monitoring does not scale against machine-speed attacks.

Future-proofing requires treating lifecycle management as part of the security architecture: secure onboarding, credential rotation, software updates, and eventual decommissioning. Getting this right at the chipset and module level reduces operational complexity and risk across decades of deployment.

An Ecosystem Responsibility

Finally, the panel emphasized that secure, resilient grid modernization is not something utilities can solve alone. It requires early and ongoing collaboration across device vendors, chipset and module providers, private wireless operators, and cybersecurity firms. Security expectations must be defined upfront and enforced consistently across the ecosystem.

As intelligence becomes ambient, embedded throughout the grid, success will be measured not by how advanced the models are, but by how safely, transparently, and reliably systems operate over time. Utilities that invest now in secure-by-design devices, Private Mobile Networks, and Zero Trust architectures will be best positioned to modernize critical infrastructure while preserving trust and operational resilience.

Download also our recent white papers on Edge AI IoT, and on IoT security (Most critical IoT threats and practical defenses, and Top security features in today’s cellular IoT modules)