A Comprehensive Study of the Capabilities of Large Language Models for Vulnerability Detection

arXiv:2403.17218v1 Announce Type: cross
Abstract: Large Language Models (LLMs) have demonstrated great potential for code generation and other software engineering tasks. Vulnerability detection is of crucial importance to maintaining the security, integrity, and trustworthiness of software systems. Precise vulnerability detection requires reasoning about the code, making it a good case study for exploring the limits of LLMs' reasoning capabilities. Although recent work has applied LLMs to vulnerability detection using generic prompting techniques, their full capabilities for this task and the …

abstract arxiv capabilities code code generation cs.cr cs.lg cs.se detection engineering importance integrity language language models large language large language models llms making reasoning security software software engineering study systems tasks type vulnerability