Data of naturalistic driving studies have shown that conventional Collision Avoidance Systems (CAS) activate a high rate of unnecessary alarms. These alarms are associated with situations predefined as hazardous by the system algorithm, while not perceived as hazardous by the human user. This driving simulator study aimed to identify factors that cause discrepancies between human hazard perception and the risk assessment of CAS. All participants encountered situations with a braking lead vehicle that would have usually activated alarms in conventional CAS. However, drivers’ hazard perception and the intensity of their braking responses varied dependent on the outcome of the conflict that either remained or dissolved. Additionally, the predictability of the outcome influenced drivers’ hazard perception, while it had no impact on their driving behaviour.