Proceedings of the Annual Meeting of the Cognitive Science Society

Cooperative driving depends on time-critical vehicle-to-everything (V2X) alerts whose evidence is distributed, incomplete, and sometimes adversarial. A key cognitive failure is \emph{miscalibrated reliance}: decision-makers may act on invalid alerts (overtrust) or dismiss valid ones (undertrust) under time pressure. We introduce \emph{auditable trust prompts}, a decision-oriented interface that compresses verifiable cues into compact badges and an action recommendation (\textsc{rely}/\textsc{verify}/\textsc{ignore}). Prompts integrate cross-source corroboration (peer vehicles and roadside units (RSUs), plus on-board sensing such as on-board diagnostics (OBD) and global navigation satellite system (GNSS) cues), recency-weighted sender reliability, and provenance availability/verification status, while highlighting uncertainty triggers that warrant selective checking. In a human-in-the-loop alert assessment task manipulating evidence uncertainty, adversarial contamination, and congestion-driven availability stress, auditable prompts reduce unsafe overreliance on invalid alerts without a commensurate increase in missed reliance on valid ones. Behavioral logs show that verification increases mainly in hard regimes, suggesting a targeted verification pathway rather than blanket distrust. These findings suggest that auditability can serve as a usable cue for calibrated reliance, beyond backend accountability, in dynamic cooperative driving.