Xingyu Wu

While autonomous driving systems and intelligent transportation infrastructures become increasingly software-defined and network-connected, ensuring their cybersecurity has become a critical component of traffic safety. Large language models (LLMs) have recently shown promise in automating aspects of penetration testing, yet most existing approaches remain limited to simple, single-step exploits. They struggle to handle complex, multi-stage vulnerabilities that demand precise coordination, contextual reasoning, and knowledge reuse. This is particularly problematic in safety-critical domains, such as autonomous vehicles, where subtle software flaws can cascade across interdependent subsystems. In this work, we present CurriculumPT, a novel LLM-based penetration testing framework specifically designed for the security of intelligent systems. CurriculumPT combines curriculum learning and a multi-agent system to enable LLM agents to progressively acquire and apply exploitation skills across common vulnerabilities and exposures-based tasks. Through a structured progression from simple to complex vulnerabilities, agents build and refine an experience knowledge base that supports generalization to new attack surfaces without requiring model fine-tuning. We evaluate CurriculumPT on 15 real-world vulnerabilities scenarios and demonstrate that it outperforms three state-of-the-art baselines by up to 18 percentage points in exploit success rate, while achieving superior efficiency in execution time and resource usage. Our results confirm that CurriculumPT is capable of autonomous, scalable penetration testing and knowledge transfer, laying the groundwork for intelligent security auditing of modern autonomous driving systems and other cyberphysical transportation platforms.

The controller area network (CAN) bus for interconnection of electronic control units (ECUs) plays a highly important role in modern intelligent vehicles. To facilitate the CAN Bus accessing to vehicle control or diagnosis, a number of mobile APPs are designed and published by automobile manufacturers to support driving and vehicle-based social network, and some are realized through the in-vehicle infotainment (IVI) middleware. Blockchain technologies are also mature for automobiles to interact service information with the whole industry. Unfortunately, there is a serious threat of command leakage from these mobile APPs, and the reverse engineering (RE) can be exploited by hackers. Previous work has researched this threat by an automatic reverse engineering tool on both automotive android and IOS APPs. However, in such common tool, APP itself-related contexts, including the feature information of CAN Bus commands, vehicle application functions, and control diagnostic protocols, are overlooked, which might be utilized to promote the reverse engineering recall. In this paper, we propose a context-based reverse engineering approach to find deep hidden commands for further revealing security threats for blockchain-powered mobile automotive APPs. For the reverse engineering, we design a context model of four-order tensor to organize multidimensional contexts and establish a continuous updating mechanism. Based on the context model, we further develop two basic analysis algorithms, max-compute (A) and clustering (A), to perform the analysis of CAN Bus commands. Extensive experiments are conducted, and we evaluate it by two metrics, recovered ratio and correctness ratio. Experimental results and the case studied on the familiar APP Carly validate the effectiveness of our approach and reveal the threat of command leakage.