PhD Position: Enhancing Software Security with LLM Reasoning for Vulnerability Detection and Repair

This PhD position at Inria Rennes focuses on advancing software security through the use of Large Language Models (LLMs). The project, funded by the ANR PRCI SecLLM4SVD initiative, aims to develop RL-based fine-tuning techniques to enhance LLM reasoning capabilities for vulnerability detection and automated code repair in repository-level software projects. The successful candidate will investigate methods to improve LLM reliability against adversarial attacks, enabling deeper vulnerability discovery and robust patch generation while preserving code functionality. The work involves collaboration with Dr. Yuejun Guo at the Luxembourg Institute of Science and Technology, with supervision by Dr. Yufei Han at Inria. The position offers a 3-year contract starting February 1, 2027, and is based in Rennes, France.

• Develop RL-based fine-tuning techniques to improve LLM reasoning for detecting vulnerabilities in large-scale software repositories.
• Investigate methods to enhance LLM reliability against adversarial attacks, such as variable renaming or dead code insertion.
• Conduct research to uncover hidden vulnerabilities arising from cross-function, cross-file, and cross-library dependencies.
• Design and implement methodologies for automated code repair, ensuring patches eliminate vulnerabilities without breaking existing functionality.
• Collaborate with external partners (e.g., Luxembourg Institute of Science and Technology) to integrate findings into broader research initiatives.
• Publish research findings in top-tier conferences and journals on cybersecurity and AI, such as ACM CCS, Usenix Security, NeurIPS, and ICML.