Research on Design and Implementation of an Intelligent Network Asset Search System Based on LLM Agent and FOFA

The fast-paced development of the Internet and the ever-growing complexity of network infrastructure have made network asset management and security assessment an indispensable problem for every organization. Asset search and vulnerability scanning in traditional sense have limitation against increasing sizes of networks and dynamic changes to assets. To address these challenges, this paper presents a novel intelligent network asset search system based on the integration of the large language model (LLM) agent and the FOFA (Fingerprint of Full Asset) search engine, with the aim to enhance the efficiency and precision of network asset identification, classification, and security vulnerability scanning. Our work leverages the capabilities of advanced natural language processing offered by LLM agents and feeds the common sense captured by FOFA within our designed model to perform even more intelligent and contextual querying. LLM agents produce sophisticated search requests derived from ever-evolving elements within the system, allowing them to perform reconnaissance on web assets in real-time. FOFA is employed for this purpose given its vast database of devices, services, and vulnerabilities that are exposed to the internet, which improves the accuracy of the asset discovery. Additionally, the system implements a feedback loop, allowing the LLM agent to dynamically refine the search strategy depending on incoming new data, optimizing search efficiency and effectiveness. Through experiments, it is shown that the proposed system is significantly better than the traditional method in search accuracy and asset discovery speed, and also performs well in vulnerability prediction accuracy.