AI 자율주행자동차 사고 책임과 형사제재

While Korea currently lacks regulations for SAE Level 4 or higher autonomous vehicles, Germany and Japan have established a legal foundation to enable Level 4 or higher autonomous driving through revisions to their Road Traffic Act. Level 4 or higher autonomous vehicles are a prime example of physical AI, operating as an AI-based autonomous driving system. Autonomous vehicles are considered “high-risk” AI, and the legal principles of risk permitted under criminal law cannot be hastily applied. Furthermore, since risk-based criminal law leads to an expansion of state penal power, AI autonomous vehicles can accept a certain level of risk confirmed not to violate the principle of proportionality. The prerequisites and requirements for recognizing criminal liability in AI autonomous vehicle traffic accidents are not only much stricter than those for civil liability but also carry strong ethical implications. Therefore, solutions currently proposed primarily involve strengthening civil damages, expanding insurance coverage, or broadening product liability. However, relying solely on civil law solutions offers no way to prevent the wrong decision where operators of autonomous vehicles, judging the costs they bear to be lighter than the administrative oversight costs imposed by authorities, might recklessly accept potential future costs. In Germany and Japan, if autonomous vehicles cause accidents or pose specific risks, a gap in criminal law will arise. As AI approaches the point of becoming a “black box”—making decisions difficult to predict and beyond human control—criminal sanctions are necessary to address this risk. If an AI autonomous vehicle is linked to an act illegal under criminal law, security measures based on the principle of proportionality could be imposed instead of criminal penalties in the future. AI technology, a trade secret of big tech companies, should be disclosed if it poses a significant social risk and clearly provides a superior benefit for preventing such risks. Based on this, a security measure system is needed as a criminal sanction to prohibit the operation of problematic individual AI autonomous vehicles.