Securing Virtual Reality Experiences: Unveiling and Tackling Cybersickness Attacks with Explainable AI

AI systems protecting VR users can be manipulated through "adversarial attacks": The paper demonstrates that tiny, invisible changes to VR data can trick AI cybersickness detection models into failing to protect users, potentially leaving them experiencing discomfort or safety risks without intervention.

Current VR safety systems have a critical vulnerability: Deep learning models used to automatically detect and mitigate cybersickness (motion sickness in VR) are susceptible to being fooled by adversarial perturbations, meaning the protective technology itself can be compromised.

Explainable AI offers a solution for detecting attacks: The researchers propose using explainable AI (XAI) methods to create a detection framework that can identify when cybersickness detection systems are under attack, helping restore the reliability of VR safety protections.

This represents an emerging class of security threats: The paper introduces "cybersickness attacks" as a new type of VR-specific threat that didn't exist before AI-based safety systems were deployed, highlighting how protective technologies can create new vulnerabilities if not properly secured.

Responsible AI deployment requires adversarial robustness testing: The research underscores the importance of testing AI safety systems against adversarial attacks before deployment in consumer applications, particularly for systems protecting user health and safety in immersive environments.