Question

A financial services firm uses a fine-tuned model in Microsoft Foundry that is grounded on a curated dataset of internal loan-underwriting policies. Compliance requires that only a small team of model engineers can modify the fine-tuning dataset and initiate tuning runs, while a larger group of business analysts may query the deployed model but must never see or alter the grounding source or training data. As the solutions architect, what is the most appropriate way to enforce this separation?

Accepted Answer

Apply role-based access control so the model engineers hold write/tuning permissions on the training dataset and tuning jobs, while analysts are granted only inference (query) permissions on the deployed endpoint with no access to the grounding data store.

Answer

Give both groups Contributor access to the Foundry project and rely on a documented policy instructing analysts not to open the training dataset.

Answer

Store the grounding dataset in the same container as the deployed model and share a single read/write key with all users, logging access afterward for review.

Answer

Deploy two identical models—one for engineers and one for analysts—so each group works in isolation without configuring differentiated permissions.