NSF Backs $500,000 Push to Secure the AI-Powered Edge of the Internet of Things

The devices that make up the Internet of Things have quietly become the nervous system of modern life. Sensors in factories, smart meters on city streets, cameras at intersections, medical monitors in hospitals, and voice assistants in living rooms all collect, process, and exchange data at the edge of the network, far from the fortified data centers where most traditional cybersecurity defenses were designed to operate. As artificial intelligence moves into these devices, the security challenge is changing shape faster than the workforce trained to meet it. Two researchers at George Mason University have now received a substantial federal award to confront that gap, combining cutting-edge research into foundation models with a sweeping redesign of how students learn to secure the connected world.

Jianli Pan, an Associate Professor of Information Sciences and Technology in George Mason’s College of Engineering and Computing, and Md Arafin, an Assistant Professor of Cyber Security Engineering in the same college, have secured $500,000 from the U.S. National Science Foundation for a project aimed at modernizing Internet of Things and edge cybersecurity research and curricula. The funding began in September 2026 and runs through late August 2029, giving the team a three-year window to build both the science and the teaching infrastructure needed for what they describe as the embodied AI era, a period in which intelligent systems no longer merely analyze data in the cloud but act in the physical world through devices embedded in homes, workplaces, and critical infrastructure.

The central premise of the project is that foundation models, the large-scale neural networks that have transformed fields from language processing to computer vision, are now migrating onto resource-constrained devices at the network edge. This migration creates a fundamentally new security landscape. When a foundation model runs on an edge gateway, a smart camera, or an industrial controller, it must be deployed, optimized, and defended within tight limits on memory, computation, and energy. Those constraints open attack surfaces that conventional cybersecurity curricula rarely address, from model extraction and adversarial manipulation to compromises of the underlying system software that hosts the model. Pan and Arafin intend to study these challenges directly while translating their findings into classroom-ready material.

The research component of the project focuses on emerging foundation models for the edge, examining how such models can be deployed and optimized on IoT platforms without sacrificing security. The investigators plan to develop innovative research and learning materials that address the full lifecycle of edge-deployed intelligence, from initial deployment decisions through performance optimization to the design of protective mechanisms. This dual emphasis on capability and security reflects a growing recognition in the field that AI systems at the edge cannot be treated as ordinary software. Their statistical nature, their dependence on large training datasets, and their integration with physical sensors all create vulnerabilities that demand new analytical tools and new system architectures.

Education sits at the heart of the effort. The researchers will teach emerging foundation-model concepts to students at different educational levels, an unusually broad ambition that spans the pipeline from early coursework to advanced graduate research. To do this, they will create structured modules that give students advanced research and experiential learning opportunities in foundation-model-enabled cyber analytics at the edge. Experiential learning, in which students confront real systems and real constraints rather than abstract exercises, has become a priority for cybersecurity educators who argue that the field’s rapid evolution leaves traditional lecture-based instruction perpetually behind the threats it must counter.

A distinctive element of the project is its attention to secure kernels and system architectures for edge-deployed foundation models. The team will develop innovative training materials on this topic, giving students a grounding in how the lowest layers of an edge system can be hardened to protect the intelligent services running above them. A secure kernel provides a trusted foundation on which higher-level AI workloads can operate, isolating them from tampering and limiting the damage an attacker can do if a device is compromised. Teaching this material alongside foundation-model deployment is intended to produce engineers who understand the entire stack, from silicon-level protections to the behavior of a neural network responding to sensor input.

The project’s outcomes are designed to reach far beyond a single university. Pan and Arafin plan to disseminate their results through a new textbook dedicated to the subject, a resource that could quickly become a reference for institutions racing to add edge AI security to their programs. The dissemination plan also includes hands-on laboratories supported by complementary virtual-machine-based tools, which allow students to experiment with edge deployments and security configurations without needing specialized hardware in every classroom. Virtualized laboratories have proven especially valuable in cybersecurity education because they permit safe, repeatable experimentation with attacks and defenses that would be impractical or irresponsible to stage on live networks.

Training modules and workshops will extend the project’s reach to audiences beyond enrolled students, offering pathways for practitioners and educators to update their skills as foundation models spread through the IoT ecosystem. A cybersecurity competition rounds out the plan, giving participants a high-stakes arena in which to apply what they have learned. Competitions have long served as a proving ground for cybersecurity talent, and anchoring one in the emerging domain of edge AI security could help establish the problem sets and defensive techniques that the next generation of practitioners will need to master.

The strategic significance of the award lies in its timing. Industry analysts project tens of billions of connected devices in the coming years, and the arrival of embodied AI, in which models perceive and act through physical hardware, multiplies both the opportunities and the risks. Securing that ecosystem requires a workforce fluent in two domains that have historically been taught separately: distributed systems and networking on one side, and machine learning on the other. By fusing them into a single research and teaching program, Pan and Arafin are betting that the most effective defense against future attacks on intelligent edge infrastructure will be built in the classroom as much as in the laboratory.

For George Mason University, Virginia’s largest public research university, the project reinforces a broader push into engineering and computing at scale. The institution enrolls more than 40,000 students from 130 countries and all 50 states, and its proximity to Washington, D.C. places it near a dense concentration of federal agencies and defense contractors with acute interests in IoT and edge security. Over the three-year funding period, the award is expected to produce research findings, educational materials, and trained graduates whose influence may extend well past the university’s campus, shaping how an entire generation of engineers designs and defends the intelligent, connected devices that are rapidly becoming inseparable from everyday life.

Subject of Research: NSF-funded modernization of IoT and edge cybersecurity research and education using foundation models

Article Title: Pan & Arafin modernizing IoT/Edge cybersecurity research & curricula

Article References: Pan & Arafin modernizing IoT/Edge cybersecurity research & curricula. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: Internet of Things, edge computing, cybersecurity, foundation models, embodied AI, National Science Foundation, George Mason University, curriculum development, secure kernels, experiential learning, edge AI security, workforce training