Cloud-Native Adversarial Machine Learning: A Serverless Cybersecurity Architecture for Neutralizing Prompt Injections in Large Language Models
The integration of Large Language Models into enterprise network architectures has introduced severe cybersecurity vulnerabilities, most notably adversarial prompt injection and zero-day data extraction attacks. Traditional network security protocols are fundamentally ill-equipped to defend against non-deterministic generative models, where malicious payloads are obfuscated within natural language. This paper proposes a cloud-native architectural framework utilizing Amazon Web Services to construct a real-time, serverless cybersecurity pipeline for Large Language Models. By deploying asynchronous Python middleware integrated with secondary classification algorithms and contextual vector boundaries, the proposed system intercepts, analyzes, and neutralizes malicious prompts in transit before they reach the primary inference engine. Preliminary architectural evaluations demonstrate that decoupling the security filtering layer from the foundational model significantly reduces the success rate of adversarial attacks without introducing prohibitive network latency. This methodology provides cybersecurity researchers and network scientists with a deterministic, highly scalable technological foundation for securing applied Artificial Intelligence in commercial environments.