Network Protocol Security Evaluation via LLM-Enhanced Fuzzing in Extended ProFuzzBench
摘要
Network protocol is of paramount importance as it serves as the bridge for secure communication within and between networks, and protocol security is vital for ensuring the integrity, confidentiality, and availability of data transmission. Nonetheless, protocol testing is a challenging task. we propose LLMFuzz, a novel LLM-based network protocol testing approach, and extend ProFuzzBench by integrating three advanced fuzzers, AFLNet Legion, ChatAFL, and LLMFuzz—along with three protocols (HTTP, MQTT, and Modbus). Experimental results demonstrate that our newly proposed and incorporated LLMFuzz outperforms the baseline fuzzer AFLNet and state-of-the-art fuzzer ChatAFL across multiple network protocols and evaluation metrics.