Abstract: The continuous emergence of software vulnerabilities presents significant threats across various industries, with their increasing complexity posing formidable challenges to the current cybersecurity landscape. As large language models (LLMs) continue to rise in prominence, the challenge of harnessing their capabilities to improve automated vulnerability detection and overcome existing technical limitations has become a critical issue. Firstly, this paper summarizes the key characteristics and limitations of traditional automated vulnerability detection tools. It then reviews research on the application of deep learning in vulnerability detection, emphasizing the theoretical and practical significance of utilizing LLMs for automated detection. Finally, the paper outlines potential research directions, challenges, and opportunities in the field of LLM-powered automated vulnerability detection.