This paper studies the effect of negative safety-related information on autonomous vehicle (AV) mode choice in Singapore. The authors implemented a dynamic online survey with half of the subjects receiving negative safety information about AV via a randomized control trial. The authors use a video of Tesla’s fatal crash under autopilot as an approximation of the negative safety information of AV. The authors test the impact of watching this video and its interaction with prior knowledge of AV technology and safety on changes in mode choice stated preference. The survey integrates revealed preference (RP) and stated preference (SP) scenarios, by generating the SP data directly based on the reported RP values. A nested logit modeling framework is built to jointly estimate the RP and SP survey data (N~2000). The modeling results reveal that the information effect of the crash video varies largely by sociodemographic group. Subjects with high income and with high education are less influenced by the information, but in different ways. High income respondents are more tractable when receiving new information compared with their prior knowledge of AV safety, but highly educated respondents are more influenced if the information is consistent with and an enhancement of their prior knowledge. The fading effect—if the impact of negative information persists over time—is also studied. The immediate effect on a young, high-income, full-time employee with graduated degree fades easily, while the impact on the subjects in low-income low-education and older groups may persist over time.