An organization is deploying several generative AI workloads by using Microsoft Foundry. Each
workload must meet different requirements related to data governance, task specialization, and
operational cost control.
The organization requires models that meet the following requirements:
- Model behavior aligns with the task being performed.
- Data handling aligns with internal governance policies.
- Operational complexity and cost are justified by workload needs.
You need to select the foundation model options that meet the requirements.
Which three models can you select? Each correct answer presents a complete solution. Choose
three.
NOTE: Each correct selection is worth one point.
Explanation:
[B] Operational complexity and cost are justified by workload needs. In a scenario where a single generative AI workload is deployed via Microsoft Foundry and the associated high operational complexity and costs are already justified, utilizing the largest available model (e.g., GPT-4 class models) is an appropriate strategy to simplify operational management. This approach aligns with a "production-first" or high-performance mindset where, instead of managing multiple smaller, specialized models, a single powerful model provides the necessary reasoning capabilities, accuracy, and broad applicability for complex, high-stakes tasks, reducing the overhead of engineering, fine-tuning, and maintaining several smaller models. [C] Model behavior aligns with the task being performed In Microsoft Azure AI Foundry, the most appropriate model to minimize usage costs while maintaining task alignment is typically a Small Language Model (SLM) from the Phi family or a specialized "mini" partner model. Top Recommended Small Models for Cost-Efficiency These models are designed for high efficiency and lower latency, making them ideal for specific, well-defined tasks. [E] Data handling aligns with internal governance policies In a Microsoft Foundry (Azure AI Foundry) setup, the Azure OpenAI Service models are the appropriate choice for processing regulated business data due to their built-in enterprise governance and security controls. Unlike standard public models, these "Direct Models" are hosted within your Azure tenant and adhere to strict organizational policies. Enterprise Governance Controls For workloads involving regulated data, the following governance features in Microsoft Foundry ensure compliance: Foundry Control Plane: Centralizes management, observability, and compliance enforcement for all models and agents in one interface. Azure Policy Integration: Allows administrators to enforce specific model usage, restrict deployments to approved regions, and apply predefined security configurations. Microsoft Purview Integration: Extends data security posture management (DSPM) to AI workloads, enabling sensitive data discovery, classification, and data loss prevention (DLP) across user prompts and model responses. Microsoft Entra Agent ID: Automatically assigns a unique identity to every AI agent, enabling granular, role-based access control (RBAC) and auditability of model interactions. Azure AI Content Safety: Provides configurable filters to block harmful or regulated content and includes "protected material detection" to prevent copyright risks. Reference: https://azure.microsoft.com/en-us/products/ai-foundry/modelshttps://learn.microsoft.com/en-us/azure/foundry/responsible-ai/openai/data-privacy