The ongoing development of acoustic event recognition methods and Sound Event Detection (SED) systems creates new opportunities to support the safety of people with hearing loss. This paper establishes a behavioral reference point for the design of such systems by measuring, in a controlled field experiment, how people respond to hazardous acoustic signals without assistance. Sixty participants took part, in two groups of thirty, one reporting no hearing loss and one with hearing loss, and each completed 20 stimulus exposures, giving 1200 trials in total. Two outcomes were analyzed separately: whether a response was recorded at all, and, conditional on a response, how long it took. In the hearing-loss group, 49.2 % of trials produced no recorded response, compared with 0.0 % in the comparison group, a difference of 49.2 percentage points (95 % confidence interval [46.0, 52.3]) amounting to complete separation between the groups, as every participant with hearing loss failed to respond to at least 30 % of signals while no comparison participant missed any. Evaluated against the standardized 5500 ms observation window, the no-response rate was 52.3 %. Among trials with a recorded response, the median reaction time was 502 ms in the comparison group and 670 ms in the hearing-loss group. These results indicate that the principal design challenge is not only the speed of the response but the frequency with which a hazardous signal produces no behavioral response at all. On this basis the paper formulates provisional design hypotheses for the AI EAR SUPPORT system, covering on-device inference and multi-channel notification through a smartphone and a smartwatch. No artificial intelligence model, notification interface, or assisted user response was tested in this study, and the hypotheses therefore require empirical validation.