Trust in the wild: Multi-modal safety perceptions in naturalistic AV deployment

Public acceptance remains a major barrier to autonomous vehicle (AV) deployment, yet most perception research relies on hypothetical surveys or small controlled experiments with AV-naive respondents. This cross-sectional study surveys 811 San Francisco Bay Area residents with real-world AV exposure (376 robotaxi passengers and 435 non-passengers: pedestrians, cyclists, and drivers) in one of the world’s most mature deployment environments. We ask whether trust levels among AV-exposed residents differ from national baselines, whether trust varies by interaction mode, and whether experience is associated with differential recalibration of specific safety concerns. We find that 61% trust AVs to be safe, 4.7 times the 13% national rate. Trust varies sharply by mode: passengers report 82% trust versus 43-45% among non-passengers. Even non-riders far exceed national levels, consistent with ambient exposure beyond ridership alone contributing to higher trust, though residential selection cannot be ruled out. Concern about system malfunction is significantly lower among passengers (45%) than pedestrians and cyclists (63%), consistent with recalibration through direct performance feedback, whereas concern about unpredictable human behavior remains constant across all modes ( ∼ 55%). These patterns suggest that experience is associated with attenuation of technology-related concerns while human-behavior concerns persist: a divergence with implications for communication strategies and infrastructure design as AV deployment expands.