AI-Driven Key Length Optimization for Energy- Efficient RSA in IoT Networks

The rapid proliferation of Internet of Things (IoT) ecosystems has intensified the demand for secure communication among resource-constrained devices, yet traditional RSA cryptography imposes prohibitive computational and energy costs on low-power nodes as larger key sizes are required for adequate security. This paper introduces an AI-driven adaptive framework that dynamically optimizes RSA key lengths to achieve an efficient security–energy trade-off in IoT environments. The approach formulates the problem as a constrained multiobjective optimization and employs a lightweight reinforcement learning agent, trained offline and deployed on-device, to select the most suitable key length (2048-bit or 3072-bit) based on real-time contextual factors including battery level, system load, and dynamic threat exposure. Extensive evaluations were performed through high-fidelity cycle-accurate simulations calibrated to real ESP32 and ARM Cortex-M4 hardware profiles. Results demonstrate that the adaptive system reduces average energy consumption by 58% and computational latency by 58% compared to a static 3072-bit configuration, while delivering a comparable security level (average 115-bit equivalent). Pareto frontier analysis confirms that intelligent switching between 2048-bit and 3072-bit keys yields the most practical balance for IoT deployments. These findings highlight how context-aware cryptographic adaptation can significantly enhance the energy efficiency and operational sustainability of secure IoT systems without sacrificing required security guarantees.