Non-Repudiation in Decentralized Wireless Networks in the Age of AI: A Comprehensive Review

Non-repudiation—the guarantee that a party cannot credibly deny a performed action—has become a first-order security primitive for decentralized wireless ecosystems that now span mobile ad hoc networks (MANETs), massive Internet of Things (IoT), vehicular networks, and emerging sixth-generation (6G) architectures. Artificial intelligence (AI) introduces dual-use dynamics: generative models empower deepfake and adversarial spoofing attacks, yet AI also enables intelligent evidence collection, anomaly detection, and forensic explainability. This survey provides a synthesis of state-of-the-art non-repudiation mechanisms in this evolving landscape. We (i) review historical approaches to non-repudiation, (ii) catalog adversarial threats from AI and AI-driven defensive mechanisms, (iii) analyze non-repudiation in federated learning, edge AI, and 6G service-based architectures, (iv) review post-quantum and lightweight signcryption schemes that make cryptographic evidence affordable for constrained devices, and (v) introduce a new four-dimensional taxonomy (trust model, resource overhead, scalability, evidence strength). Comparative tables and figures quantify latency, energy, and signature size across representative schemes. We conclude with open challenges—including deepfake-aware authentication, privacy-aware logging, and quantum-resilient evidence chains—and propose directions for standards bodies and industry adoption.