In unavoidable traffic accidents, autonomous vehicles (AVs) face the dilemma of protecting either the passenger(s) or third parties. Recent studies show that most people prefer AVs following a utilitarian approach by minimizing total harm. At the same time, however, they would adopt an AV only if it prioritizes the passenger(s), i.e. themselves. As AVs exhibit a lower accident risk in the first place, a regulator therefore faces a trade-off: the harm-minimizing behavior of AVs (ex post efficiency) hampers the willingness to adopt them (ex ante efficiency). Using a gametheoretic model, we analyze how the second-best optimal level of AV passenger protection depends on (i) the AV safety advantage, (ii) the intensity of drivers’ social preferences, and (iii) their reluctance to adopt AVs. A higher AV safety advantage may either increase or decrease the second best optimal level of passenger protection.