A wearable hip airbag must recognize a balance disturbance early enough to inflate before the body reaches the ground. Unexpected postural perturbations evoke a cortical response recordable behind the ear. We developed a detector that uses one mastoid electroencephalography (EEG) channel and a network small enough for a microcontroller. EEG and platform acceleration were recorded from 40 young and older adults during 34 translations each; no falls occurred. A 1570-parameter EEGNet was evaluated with leave-one-subject-out cross-validation, running continuously over entire recordings, with the accelerometer only gating classification. The detector recognized 94.1% of the unexpected perturbations, rejected 92.7% of the unlabeled movements that reached the classifier and decided a median of 364 ms after the detected acceleration onset, within 400 ms for 90.9% of the perturbations. On an ESP32-S3 microcontroller, one inference took about 25 ms, leaving 78.4% of the decisions within 400 ms. Starting the analysis window 60 ms before the detected onset instead of 80 ms after it raised the area under the receiver operating characteristic curve from 0.898 to 0.962 without adding latency. This leaves time for airbag inflation before impact in typical falls; the main barrier, 37–46 false alarms per hour, sets the target for inertial confirmation.