Modern practices for assessing the safety of highly automated vehicles (HAVs) rely heavily on expert judgment, which increases both the labor intensity and the time required to conduct such assessments. Additionally, the use of expert methods makes safety evaluations susceptible to human error. This raises a significant issue in the design of HAVs: if the safety assessment results are negative, redevelopment becomes necessary, leading to additional resource expenditure. Another challenge is the inability to formally demonstrate the correspondence between the safety conclusions drawn by experts and the actual conditions in road scenarios. Complicating matters further, there are often discrepancies in assessments among experts themselves. It is hypothesized that artificial intelligence (AI) methods could assist experts in making more accurate, stable, and faster safety assessments. In this paper, the authors analyze various AI methods applicable to safety evaluation, including their own method based on Bayesian networks. They also compare these methods with the current expert-based approach used by functional safety specialists and draw conclusions regarding the acceptability of each method.