Leveraging Trustworthy AI for Automotive Security in Multi-Domain Operations: Towards a Responsive Human-AI Multi-Domain Task Force for Cyber Social Security

Connected autonomous vehicles face multiple simultaneous AI-based attacks: The research identifies that adversarial machine learning can compromise multiple in-vehicle systems at once (such as intrusion detection and traffic sign recognition), creating compounded security risks for smart cities and autonomous vehicles that go beyond single-point failures.

Machine learning model design choices directly impact vulnerability windows: The study found that specific hyperparameters in decision tree-based models—particularly the number of trees and boosting rounds—significantly affect how quickly attackers can execute black-box attacks, meaning that technical configuration decisions have real security consequences.

Different AI models have different security profiles: Random Forest and Gradient Boosting models were found to be more sensitive to parameter changes than Extreme Gradient Boosting, suggesting that organizations cannot assume all machine learning approaches offer equivalent security levels when deployed in critical infrastructure.

Defensive AI training requires careful timing considerations: The research analyzes how long it takes to implement adversarial training (a defensive technique), highlighting that security teams need to understand the "window of opportunity" attackers have before defenses can be deployed—a practical constraint for protecting autonomous systems.

Responsible AI requires cross-domain security thinking: The framework emphasizes that trustworthy AI in civilian infrastructure like autonomous vehicles must be designed with military-grade multi-domain threat awareness, suggesting that responsible AI deployment in critical systems demands coordination between civilian and security sectors.