Question 730
Following an approved migration, the team revisits: A Kubernetes AI inference service has flat pod reachability to finance databases. What test result is most decisive?
Explanation
Correct Answer
Inference pods reach approved model dependencies but cannot directly connect to finance data stores. It validates the recommended action: Combine AI runtime safeguards with Kubernetes workload segmentation at the appropriate egress boundary. The design principle is: Model-content inspection and east-west network isolation mitigate different attack surfaces.
Incorrect Answers
A controlled adversarial prompt suite produces findings before production exposure. That observation supports a DIFFERENT situation: A new enterprise chatbot model has not been assessed for prompt injection before production launch. Here the selected design is: Combine AI runtime safeguards with Kubernetes workload segmentation at the appropriate egress boundary.
Reporting distinguishes recognized sanctioned AI apps from unauthorized new AI destinations. That observation supports a DIFFERENT situation: Employees use sanctioned and unknown generative AI websites, and security wants visibility across both. Here the selected design is: Combine AI runtime safeguards with Kubernetes workload segmentation at the appropriate egress boundary.
A synthetic customer dataset upload is blocked even though the base application is sanctioned. That observation supports a DIFFERENT situation: An employee AI browser plugin is labeled approved but exports internal customer data to a new endpoint. Here the selected design is: Combine AI runtime safeguards with Kubernetes workload segmentation at the appropriate egress boundary.