Published Paper Details:

The proliferation of latency-sensitive, mission-critical applications has exposed the limitations of conventional BGP routing in meeting stringent service-level agreements (SLAs) within modern Software-Defined Wide Area Networks (SD-WANs). This review synthesizes a decade of research on artificial intelligence (AI) techniques applied to BGP route optimization in SLA-driven SD-WAN architectures. We examine supervised learning, deep reinforcement learning, multi-agent systems, and graph neural networks, alongside predictive analytics for proactive SLA maintenance. Experimental evidence from both simulation and real-world testbeds indicates that AI-augmented approaches improve SLA compliance by 10-22 percentage points over default BGP and 5-16 points over heuristic traffic engineering, while reducing route churn and operational costs. Key challenges include ensuring trustworthy AI, handling cross-domain optimization, safeguarding against adversarial threats, and standardizing interoperability. Future research directions emphasize explainable AI, federated optimization, carbon-aware routing, and integration with intent-based networking. This review provides both a comprehensive state-of-the-art summary and a forward-looking agenda for researchers and practitioners in AI-driven network control.