Adversarial Attacks on Intrusion Detection Systems in In-Vehicle Networks of Connected and Autonomous Vehicles.

Area of Science:

• Cybersecurity
• Machine Learning
• Automotive Engineering

Background:

• Connected and autonomous vehicles (CAVs) rely on machine learning (ML) for advanced functionalities.
• ML-based intrusion detection systems (IDSs) in in-vehicle networks (IVNs) are critical for security.
• Adversarial attacks pose a significant threat to the reliability of these systems.

Purpose of the Study:

• To investigate the vulnerability of ML-based IDSs in IVNs to adversarial attacks.
• To assess the susceptibility of IDSs to manipulation, particularly given the simpler nature of IVN data compared to perception models.
• To propose and evaluate a novel adversarial attack method targeting IVN IDSs.

Main Methods:

• Developed a black-box adversarial attack using a substitute IDS trained on onboard diagnostic port data.
• Simulated attacks under realistic IVN traffic constraints.
• Evaluated the attack's effectiveness against a baseline IDS and a state-of-the-art model (MTH-IDS).

Main Results:

• Demonstrated substantial vulnerability of both IDS models to the proposed attack.
• Achieved significant reductions in F1 scores, from 95% to 38% for the baseline and 97% to 79% for MTH-IDS.
• Found that inducing false alarms was a particularly effective adversarial strategy, eroding trust.

Conclusions:

• IVN-based IDSs, despite their simplicity, exhibit critical vulnerabilities to adversarial manipulation.
• These vulnerabilities pose a threat to vehicle safety and user trust.
• Careful consideration is needed in developing IVN IDSs and responding to their alerts.