The fast proliferation of the Internet of Things (IoT) networks requires lightweight and yet powerful cryptographic tools to ensure secure communication within resource-limiting settings. This work presents a novel construction of the \(8 \times 8\) S-box using the generator matrix of the [15, 11] linear block code. To test its applicability in practice, the S-box becomes a part of an IoT authentication system that combines PIN verification, a one-time password (OTP) as fail-and-save mechanism, and a post-quantum learning-with-errors (LWE) challenge-response protocol. Moreover, an artificial intelligence-based analysis tool is created to categorize and evaluate the cryptographic strength of the proposed S-box. Experimental findings prove outstanding performance: Naïve Bayes, Decision Tree, Random Forest, BERT, RoBERTa, and LLaMA demonstrate ideal accuracy, precision, recall, and F1-scores which further confirm the reliability, scalability, and reproducibility of the suggested evaluation framework. These findings confirm that the constructed S-box not only achieves strong cryptographic properties but also provides a validated, AI-assisted way of improving confidentiality and reliability in telecommunication systems based on the internet of things.
