HUMAN-CENTRIC SECURITY ASSURANCE IN AI-AUGMENTED SOFTWARE ENGINEERING ECOSYSTEMS: EXPLORING DEVELOPER STRATEGIES FOR VALIDATING LLM-GENERATED CODE ARTIFACTS

Abstract

The rapid adoption of Large Language Models (LLMs) has transformed software engineering by enabling developers to automate code generation, debugging, testing, and documentation. While these technologies enhance productivity, they also introduce security vulnerabilities, inaccurate logic, dependency risks, and governance challenges that require effective human oversight. This study explores how software developers in Pakistan validate LLM-generated code artifacts and implement human-centric security assurance within AI-augmented software engineering ecosystems. Guided by an interpretivist philosophy, the study employed an exploratory qualitative design and conducted semi-structured interviews with 17 software professionals working across software houses, IT export companies, FinTech organizations, technology startups, multinational firms, and public-sector digital organizations. Data were analyzed using Reflexive Thematic Analysis. The findings identified six interconnected themes highlighting the importance of human judgment, multi-layered validation, security awareness, organizational governance, continuous human-AI learning, and Human-Centric Security Assurance as a socio-technical capability. The study extends Human-Centered AI Theory by demonstrating that secure AI-assisted software engineering depends on the interaction of developer expertise, structured validation practices, and organizational governance rather than AI capabilities alone. The findings provide theoretical insights and practical recommendations for strengthening secure AI adoption within Pakistan's evolving software industry and other emerging digital economies.