Search for a command to run...
A Single Neuron Is Sufficient to Bypass Safety Alignment in Large Language Models