Model Denial of Service (DoS)

Explanation

This Model Denial of Service control measures average response time under load, verifying the AI service stays responsive when traffic is heavy rather than slowing to the point of effective unavailability — a resilience requirement under RBI Safety/Resilience, NIST MANAGE 2.1/3.3 and ISO Clause 6.1.2. Latency that balloons under load is itself a denial of service: a fraud-screening or chat assistant that takes too long to answer fails customers just as surely as one that is offline. The metric is Average Response Time Under Load, with the proposed expectation that it remains within the capacity SLA and that the system auto-scales or alerts at 80% utilisation. No calculation method is specified, so measure it as mean (and ideally tail) response latency observed while the system is under elevated request volume, compared against the latency commitments in the capacity SLA. To implement, instrument the inference path with end-to-end latency timers, run load tests to characterise behaviour near capacity, and tie autoscaling and queue management to the 80% utilisation trigger; log latency distributions and scaling events as evidence. When utilisation reaches 80% or response time breaches the SLA, the control alerts and scales out, escalating persistent slowdowns to capacity and architecture review.