Accelerated DevSecOps Automation through Large Language Models and Cloud-Native Security for Vulnerability Management

Abstract

The rapid adoption of cloud-native development has increased software delivery speed while simultaneously introducing complex security challenges across containers, microservices, APIs, infrastructure-as-code, dependencies, and continuous integration and continuous deployment pipelines. Conventional vulnerability management approaches often depend on periodic scanning, manually interpreted findings, and fragmented security tools, creating delays between vulnerability discovery, prioritization, remediation, and verification. Large Language Models (LLMs) offer opportunities to enhance DevSecOps by interpreting security findings, correlating vulnerability information with application context, generating remediation recommendations, assisting developers with secure code changes, and automating repetitive security workflows. This paper examines the integration of LLM capabilities with cloud-native security practices to accelerate vulnerability management throughout the software development lifecycle. The proposed approach combines continuous security monitoring, automated vulnerability classification, contextual risk prioritization, LLM-assisted remediation, policy enforcement, and automated verification within DevSecOps pipelines. The research methodology uses a structured literature review, conceptual architecture development, scenario-based experimentation, and quantitative evaluation. Key evaluation dimensions include vulnerability detection accuracy, prioritization quality, remediation time, false-positive reduction, developer effort, pipeline performance, and remediation verification accuracy. The study also addresses security risks associated with LLM adoption, including hallucinations, prompt injection, insecure generated code, data leakage, and excessive automation. The research positions LLMs as controlled intelligence components within cloud-native security architectures rather than unrestricted autonomous decision-makers.