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 …

arxiv capabilities code cs.cr cs.lg cs.se detection engineering great integrity language language models large llms making reasoning security software software engineering software systems study systems trustworthiness vulnerability vulnerability detection