EP03 - Nobody Told You This Microsoft Foundry Security Model Checklist

Описание: Secure, compliant AI endpoints that keep your tenant data inside Microsoft Foundry — Microsoft Foundry guidance for SharePoint/Power Platform builders. Microsoft Foundry is the primary keyword and the benefit: choose safe models, avoid data exfiltration, and deploy endpoints with confidence.

Who this is for: Enterprise architects, SharePoint/Power Platform developers, .NET integrators, and Azure admins building AI-enabled apps in Microsoft 365.

What you’ll learn:
The single critical flag that tells you if a model is hosted "Direct from Azure" or by a third party.
Why Azure-provided models keep your data inside your tenant and what that means for compliance.
The fallout from using partner/community models (e.g., Anthropic) and how data can leave your tenant.
The Foundry hub → services → project architecture and why hubs matter for service scoping.
How Azure OpenAI is being folded into Microsoft Foundry and the migration implications.
Practical checklist questions to ask before deploying any model endpoint in Foundry.
How to spot the filter and model catalog indicators that guarantee MS-hosted copies.
What to document in your security review to keep auditors and CISO teams happy.

Series: Power Agents Series • EP03

What this video covers in one line: Build safe endpoints in Microsoft Foundry so your LLM calls, data at rest, and transit remain inside your Azure tenant — avoid hidden third-party hosting.

Chapters:
0:00 - Intro: Why this matters now
0:19 - The Foundry vs Azure OpenAI pivot
0:48 - Azure-provided models explained
1:18 - What "provided by Azure" actually means
1:41 - Risk: Non-Azure models and data exfiltration
2:58 - Foundry architecture: service, hubs, projects
3:54 - History: Azure OpenAI to Foundry transition
4:39 - Safe boundary: tenant vs internet
5:03 - How OpenAI models are hosted in Azure
5:28 - Model catalog categories explained
6:01 - Partner/community models vs Azure hosted
6:35 - Anthropic example: what changes
7:22 - Live demo tease: existing tenant view
8:15 - The catalog filter you must enable
8:44 - Why "Direct from Azure" filter is critical
9:28 - What to check on partner model pages
10:00 - Final notes & next video preview
10:23 - Outro: setup from scratch coming next

Key takeaways / Actionable checklist:
Always enable the "Direct from Azure" filter when picking models.
Treat any model flagged as "provided by partners/community" as a potential external-host risk.
Document model provider, hosting guarantees (at-rest, in-transit encryption), and data usage T&Cs before deployment.
Use Foundry hubs to centralize services (Blob, Search, Cosmos DB, Key Vault) for consistent governance.
Plan migration from Azure OpenAI resources to Foundry-managed endpoints — test endpoints before cutover.

#Microsoft365 #AIAgents #PowerAgents #PowerPlatform #AzureAIFoundry #MicrosoftFoundry #AzureOpenAI #LLMsecurity #EnterpriseAI #SharePointDev #AzureSecurity #AIGovernance #ModelSelection #EndpointSecurity