CYRUS’SCloud CornerTECHNOLOGY · TRUST · PERSPECTIVE
Home / AI / Microsoft Azure AI

Microsoft Azure AI

Microsoft Foundry is the current platform name for the developer experience previously called Azure AI Foundry. Keep product naming separate from the identity, data, and application controls that make an AI system reliable.

Azure cloud AI illustrated with Azure OpenAI, Azure Machine Learning, Microsoft Foundry, Responsible AI
Azure cloud AI — an illustrated overview of the topics on this page.

Content reviewed: October 25, 2026

Microsoft Foundry

Foundry supports building and operating AI applications with models, agents, evaluation, and related tooling. Availability and behavior vary by model, deployment type, region, and feature status. Verify each selected service rather than assuming every catalog item has identical controls.

Azure Machine Learning and search

Azure Machine Learning serves custom machine-learning development and operations. Azure AI Search can support retrieval architectures. A retrieval index does not replace source authorization: ensure the application enforces document access and propagates permission changes.

A governed Microsoft AI application: Identity — Entra and scoped service access; Data and retrieval — Preserve document permissions; Model application — Selected Foundry deployment; Operating review — Evaluation, monitoring and change control
A governed Microsoft AI application · Conceptual diagram; see the worked example and references.

Identity and tool boundaries

  • Use Microsoft Entra ID and scoped managed identities where supported.
  • Authorize access outside the model and validate tool arguments in application code.
  • Keep high-impact actions behind an appropriately reviewed execution path.
  • Review network exposure, logging, retention, and encryption for the deployed configuration.
  • Test quality, prompt injection, access violations, and recovery before expansion.

Evaluation and oversight

Maintain a representative evaluation set and review failures before changing models, prompts, retrieval, or tools. Record the deployed versions and configuration. Responsible-AI guidance and content filters support a control program; they do not certify the application or guarantee correctness.

Sources & further reading