Abstract

To distribute traffic across microservices and backend systems, modern cloud-native infrastructures primarily rely on load balancers and ingress controllers. Although the main purpose of these components is to increase availability and scalability, they are becoming increasingly important control points in network architectures, making them appealing targets for cyberattacks. Load balancers are frequently viewed as performance-oriented networking components rather than active security enforcement mechanisms, despite their strategic position. To support advanced cybersecurity capabilities, this thesis explores ways to improve load-balancing architectures. The study assesses the efficacy of AI-assisted Web Application Firewalls (WAFs) in identifying real-time threats, investigates the integration of machine learning techniques for proactive vulnerability discovery, and examines protocol-level security issues arising from contemporary transport protocols such as HTTP/3 and QUIC. The study also examines how load-balancing architectures can apply Zero Trust security principles to limit lateral movement in microservice environments and reduce implicit trust. The thesis proposes several defensive improvements for contemporary load balancing systems, drawing on experimental evaluation, architectural design, and security analysis. These include identity-based traffic routing aligned with Zero Trust principles, protocol-aware mitigation techniques for QUIC-specific attacks, and machine-learning-based anomaly detection using load balancer telemetry. The results show that load balancers can be made much more effective as security enforcement points in distributed cloud environments by incorporating AI-driven detection mechanisms and protocol-aware defenses.

Keywords