Road tunnels represent one of the most complex and safety–critical segments of road infrastructure, as they involve restricted geometry, reduced visibility, and limited opportunities for evasive maneuvers. Inadequate lighting, poor ventilation, and insufficient signaling systems significantly increase the likelihood of traffic incidents, especially when drivers are not properly informed or when technical failures occur within confined spaces. In order to analyze risk perception and assess safety in road tunnels, a comprehensive study was conducted, covering three representative scenarios corresponding to different tunnel characteristics: short urban tunnels, long motorway tunnels, and bi-directional rural tunnels. The research included 108 drivers of various profiles (professional and non-professional drivers) and 8 experts from the fields of traffic and civil engineering. The results indicate a statistically significant difference in perceived safety between the artificial intelligence model, drivers, and experts. Notably, drivers often underestimate the hazards associated with tunnel environments, particularly those related to visibility, space limitation, and emergency response conditions. Based on the findings, an integrated AI-assisted safety assessment model is proposed, which can be implemented within intelligent transportation systems and navigation platforms to provide real-time alerts, support risk mitigation, and enhance safety in tunnel environments.