Consider the following unequal error protection scenario. One special message, dubbed the "red alert" message, is required to have an extremely small probability of missed detection. The remainder of the messages must keep their average probability of error and probability of false alarm below a certain threshold. The goal then is to design a codebook that maximizes the error exponent of the red alert message while ensuring that the average probability of error and probability of false alarm go to zero as the blocklength goes to infinity. This red alert exponent has previously been characterized for any discrete memoryless channel operating at capacity and for the binary symmetric channel at any rate. In this talk, we completely characterize the optimal red alert exponent for additive white Gaussian noise channels with average power and peak energy constraints.