AI Safety Assurance for Automated Vehicles: A Survey on Research, Standardization, Regulation

Safety assurance frameworks for AI in autonomous vehicles are still underdeveloped: While AI has demonstrated superior performance compared to traditional approaches in automated driving, the paper identifies a critical gap—the safety assurance systems required to validate and certify AI in these safety-critical applications have not yet been fully established.

Research, standardization, and regulation must advance together: The paper emphasizes that because AI technology evolves rapidly, these three areas (research development, industry standards, and government regulation) are progressing in parallel. This requires coordinated effort rather than sequential development, as decisions in one area directly impact the others.

Current approaches designed for traditional systems are insufficient: Classical safety assurance methods were developed for conventional automotive technologies and cannot adequately address the unique challenges posed by AI systems, which operate with greater complexity and dynamic behavior than rule-based systems.

Interdependencies between research, standards, and regulation create open questions: The survey identifies that the interconnected nature of these three domains creates unresolved challenges that require simultaneous consideration—meaning researchers, standards bodies, and regulators must work collaboratively to address gaps rather than operating independently.

A holistic perspective is needed for responsible AI deployment: Rather than focusing on any single aspect, the paper advocates that safety experts and developers need to understand how research findings, emerging standards, and regulatory requirements interact to ensure AI systems in autonomous vehicles can be safely deployed.