Side-Channel Attack

An attack that extracts sensitive information not by directly compromising an AI system's inputs or outputs, but by analyzing indirect signals such as timing patterns, power consumption, memory access behavior, or network traffic that leak details about the system's internal state or the data it is processing. Side-channel attacks are difficult to defend against because they exploit physical and implementation characteristics rather than logical vulnerabilities in the model itself.