Model Denial of Service (DoS)

Explanation

This Model Denial of Service control monitors resource-utilisation spikes — sudden surges in CPU, GPU, memory or accelerator usage on the model-serving infrastructure — to detect overload or abuse before it degrades or topples an AI service that BFSI users rely on, consistent with RBI Safety/Resilience, NIST MANAGE 2.1/3.3 and ISO Clause 6.1.2. Spikes can stem from adversarial prompts engineered to be expensive, runaway agent loops, or legitimate demand peaks; left unmanaged they cause outages or runaway cost. The metric is Resource Utilization Spikes, with the proposed expectation of staying within the capacity SLA and auto-scaling or alerting at 80% utilisation. No calculation method is given, so track utilisation as the percentage of provisioned resource capacity consumed and watch for abnormal jumps. To implement, instrument hosts and accelerators with fine-grained resource telemetry, baseline normal usage, and detect spikes against that baseline; wire autoscaling and load-shedding to the 80% trigger and cap per-request compute to contain expensive inputs. Log utilisation series and scaling actions as evidence. When utilisation hits 80% of capacity the control alerts and scales, and an abnormal spike triggers investigation of its source — distinguishing attack from organic demand — with mitigation applied so the service stays within its capacity SLA.