Student Research Track: Hack on the Road: Can LLM-Based Driving Agents Improve Driver Response During Automated Vehicle Cyber-Attacks?

Automated vehicles (AVs) increasingly rely on connected communication systems, which make them vulnerable to cyber-attacks. During such events, drivers may not understand the situation and might need additional assistance. Although driving agents can provide warnings, large language model (LLM)-based driving agents may offer deeper explanation and follow-up support. Prior work has examined cyber-attacks in vehicles and the role of in-vehicle intelligent agents, but limited research has investigated how different agents specifically LLM-based driving agents affect driver response during AV cyber-attacks. This ongoing study proposes a driving simulator experiment comparing low-critical and high-critical cyber-attacks in three agent conditions: no agent, a rule-based driving agent, and an LLM-based driving agent. The rule-based driving agent provides short cyber-attack warnings and action instructions, while the LLM-based driving agent provides conversational explanations and follow-up question-answer support. Behavioral and subjective measures will be collected. The expected contribution has implications in the field of Human-computer interaction for the future design of driving agents according to the risk of cyber-attacks.