Reinforcement learning-based adaptive quantum-safe cryptography for DN25-compliant smart environments

Abstract

The emergence of quantum computing challenges traditional security mechanisms, particularly in heterogeneous and resource-constrained IoT and smart environments that must satisfy DN25 requirements. This work introduces a reinforcement learning-driven model for the adaptive selection and orchestration of cryptographic algorithms. Acting as an intelligent decision layer, the system observes contextual, network, and operational metrics to recommend or enforce configurations combining classical schemes, post-quantum cryptography, and Quantum Key Distribution when available. The selection problem is formulated as a Markov Decision Process with state and action spaces aligned with protocol control flows and is embedded into a security middleware with negotiation and fallback mechanisms to ensure interoperability and policy compliance without modifying application logic. Experimental results demonstrate that the model dynamically adjusts key lengths, algorithm families, and security policies according to risk and resource conditions, increasing post-quantum cryptography and Quantum Key Distribution usage by up to 33.4% and 23.9% in high-risk scenarios while favoring low-latency classical or hybrid options for less critical traffic. The system achieves success rates above 78% while maintaining stable latency and resource usage during extended operation.