Measuring and Exploiting Contextual Bias in LLM-Assisted Security Code Review

arXiv:2603.18740v2 Announce Type: replace-cross Abstract: Automated Code Review (ACR) systems integrating Large Language Models (LLMs) are increasingly adopted in software development workflows, ranging from interactive assistants to autonomous agents in CI/CD pipelines. In this paper, we study how...