Internet-of-Things (IoT) sensors have begun generating zettabytes of sensitive data, thus posing significant design challenges: limited bandwidth, insufficient energy, and security flaws. Due to their inherent trade-offs, these design challenges have not yet been simultaneously addressed. We propose a novel way out of this predicament by employing signal compression, machine learning inference, and cryptographic techniques on the IoT sensor node. Our approach not only enables the IoT system to push signal processing and decision-making to the extreme of the edge-side (i.e., the sensor node), but also solves data security and energy efficiency problems simultaneously. Experimental results on six different IoT applications indicate that relative to traditional sense-and-transmit sensors, IoT sensor energy can be reduced by 77.8× for electrocardiogram (ECG) sensor based arrhythmia detection, 808.6× for freezing of gait detection in the context of Parkinson's disease, 162.8× for neural prosthesis spike sorting, 37.6× for human activity classification, 368.4× for electroencephalogram (EEG) sensor based seizure detection, and 12.9× for chemical gas classification.