LLM-Aware Static Analysis: Adapting Program Analysis to Mixed Human/AI Codebases at Scale

This paper examines the problems and remedies of porting the static analysis methods to mixed human/AI codebases, within the context of the recently established trend to adopt large language models (LLMs) in software production. The study presents the LLM-aware static analysis, a novel method designed to address the complexity of analyzing code generated by both human and AI-based systems. This approach will enhance the quality, scalability, and security of large-scale software environments by using both conventional and AI-assisted tools in the traditional manner of analyzing the software. The research utilizes actual case studies and performance indicators to assess the efficiency of this framework in comparison with traditional static analysis methods. The major research results showed significant improvements in detection accuracy, scalability, and error reduction. The study benefits the development of program analysis techniques by offering a scalable system of mixed human/AI codebases and emphasizing its consequences in the further workflow of software development.