Denial of AI Service
Denial of AI Service is like a prankster flooding a busy restaurant with fake orders, forcing the kitchen to waste all its time cooking meals that no one will eat. Because AI systems are expensive to run and have limited capacity, these fake requests can overwhelm the system and prevent real users from getting the help they need.
Denial of AI Service (DoAIS) is an attack where an adversary exploits the high computational cost of AI inference by sending a flood of resource-intensive requests. By forcing the model to perform unnecessary, heavy calculations, the attacker exhausts the system's specialized compute resources, leading to service degradation or complete unavailability for legitimate users.
Denial of AI Service is a specialized availability attack targeting the inference pipeline of AI-enabled systems. Adversaries leverage the asymmetric cost of compute—where specific, crafted inputs trigger disproportionately high GPU/TPU utilization—to induce resource exhaustion. This attack exploits the system's reliance on expensive, finite hardware acceleration to degrade throughput or trigger a total service outage, effectively bypassing standard rate-limiting by maximizing the computational latency per request.