Devsecopsgpt: Real-Time LLM-Based Policy Enforcement During CI/CD

Abstract

Modern DevSecOps pipelines aim to integrate security practices into continuous integration and continuous delivery (CI/CD) workflows without compromising agility. However, traditional policy enforcement mechanisms struggle to keep pace with the dynamic and complex nature of modern software systems. This paper proposes DevSecOpsGPT, a novel framework leveraging large language models (LLMs) for real-time policy enforcement during CI/CD execution. DevSecOpsGPT integrates Policy-as-Code (PaC) principles into the CI/CD toolchain, orchestrating LLMs to interpret and enforce contextual security rules, detect violations, and provide just-in-time feedback. This research outlines the system architecture, enforcement strategies, and implementation methodology, and evaluates performance across various policy compliance scenarios. Our findings indicate that LLM-based enforcement enhances automation, reduces false positives, and provides adaptive learning for evolving security needs—making it a viable path toward fully autonomous secure software delivery.