Defense systems often push edge processing and sensor performance to extremes, making them strong candidates for AI—but also exposing constraints that differ sharply from commercial deployments. This talk explains the unique technical and operational challenges of deploying AI for defense sensing and proposes practical approaches to address them. We’ll cover security-driven restrictions on using open-source tooling within protected networks, the gap between raw sensor data rates and feasible on-device AI throughput and the implications of processing complex or quadrature I/Q data where standard vision pipelines do not apply. We’ll also discuss how limited access to representative training data and realistic field testing affects model validation and how responsible AI requirements become higher stakes when systems influence command-and-control decisions. Finally, we’ll examine deployment realities such as temperature, shock, vibration and packaging, and what they mean for selecting silicon and building robust edge solutions.