Question 180
While designing controls for this requirement—A multi-region AI service has strict latency targets and requests inline protection with no outage of inference during node failure—what limitation should the project not overlook?
Explanation
Correct Answer
The controlling risk is that placing one inspection point in a distant region can cause latency and availability problems. In the reported setting—a multi-region AI service has strict latency targets and requests inline protection with no outage of inference during node failure—the team needs to review supported inline AIRS deployment options, failure modes and latency budgets by region and then design supported redundant regional AI security enforcement with staged canary traffic tests. The expected safe result is that a single enforcement node failure does not bypass the agreed security policy or breach SLOs.
Incorrect Answers
That option is relevant to “A firm hosts AI workloads across cloud containers and on-prem VMs and needs consistent security outcomes despite mixed runtimes”. The design should document the corresponding investigation and practical acceptance evidence. It is not sufficient to establish the selected outcome, “Placing one inspection point in a distant region can cause latency and availability problems”, in this question.
This would be appropriate for “A security architect is asked to justify an AI protection deployment for both batch model evaluation and interactive inference”. The expected safe result is that each control generates evidence for the lifecycle phase it was selected to protect. Here the decisive requirement is “Placing one inspection point in a distant region can cause latency and availability problems”.
The alternative deals with “AI workloads in separate business units need distinct policies, but a team proposes a single unspecific runtime rule for all model APIs”. Confirm the relevant evidence—identify unit-specific sensitive data, model endpoints and identity/trust zones—before selecting design separately scoped AIRS policies by model service and risk tier, with centralized oversight. The current question instead selects “Placing one inspection point in a distant region can cause latency and availability problems”.