An Approach Based on LLMs for Forensic Threat Detection in Autonomous Systems
摘要
In a world where interaction between humans and robotic systems is increasingly common, cybersecurity is essential to protect the integrity of both. Security measures to protect humans from possible accidents or cyber-attacks on a robotic system are essential. However, after an unexpected event, whether caused by a system malfunction or the influence of a malicious actor, it must be possible to explain what happened. In this paper, we propose to combine the fields of explainability and cybersecurity by exploring the potential of an explainability system in robotics based on LLMs as a forensic tool. We performed a denial of service attack on the mobile robot’s controller. After, we analyzed the logs generated by the robot using the proposed explainability system. The results obtained show that this system can detect the presence of a potential threat affecting the robot’s behavior.