A security team wants to implement network isolation for Azure Machine Learning compute instances to prevent them from accessing the public internet while still allowing them to communicate with Azure Container Registry and Azure Blob Storage. Which Azure ML workspace configuration achieves this?

B is correct because Azure Machine Learning supports a managed virtual network feature with three isolation modes: 'Disabled,' 'Allow internet outbound,' and 'Allow only approved outbound.' In 'Allow only approved outbound' mode, all outbound traffic from compute instances is blocked by default except for Microsoft-required endpoints, and you add specific outbound rules (FQDN rules or private endpoint rules) for services like Azure Container Registry and Azure Blob Storage. This provides strong network isolation without requiring manual NSG management. A is wrong because while an NSG can block internet traffic, it requires careful management of allow-rules for Azure service dependencies and doesn't integrate as tightly with AML's managed network architecture. C is wrong because Azure Firewall is a valid but more operationally complex approach that requires deploying and managing a separate firewall resource; the AML managed virtual network achieves the same isolation more efficiently within the AML feature set. D is wrong because VNet peering connects networks together, not isolating them; additional controls would still be needed to restrict internet access.