Azure AI Foundry in 2026: 60,000 Customers, an AI Red Teaming Agent, and the Road to Enterprise Standard

Published by: JBI Training | May 2026

Category: Microsoft AI | Azure AI Foundry | Enterprise AI

At Microsoft’s 50th anniversary event, the company described Azure AI Foundry as an ‘agent factory — a modular production system designed for modern intelligence.’ The numbers behind that framing are striking: as of early 2026, more than 60,000 customers are building with Azure AI Foundry, with access to over 1,800 models and tens of thousands of organisations using the Azure AI Agent Service since its public preview in late 2024.

For technical teams evaluating where to build production AI systems on Azure, Azure AI Foundry is no longer a preview product or an Azure AI Studio rebrand. It is the Microsoft enterprise AI platform, and understanding it has become a core competency for architects and developers working in the Azure ecosystem.

What Has Changed in 2026

 

The AI Red Teaming Agent: Safety Testing at Scale

One of the most significant 2026 additions is the AI Red Teaming Agent, now in public preview. This agent systematically probes AI models and agent systems to uncover safety risks, integrating Azure AI Foundry’s evaluation systems with Microsoft Security’s PyRIT (Python Risk Identification Tool) framework. It generates comprehensive reports and tracks improvements over time — creating what Microsoft describes as an AI safety testing ecosystem that evolves alongside your system.

For enterprise teams in regulated industries, this is important. The question of how to systematically test AI agents for safety and reliability — rather than relying on ad-hoc manual testing — has not had a good answer until now. The Red Teaming Agent provides one.

The Feedback and Observability System

At the core of Azure AI Foundry’s 2026 architecture is a feedback system designed to enable continuous improvement through real-time telemetry and user feedback. This is complemented by an advanced observability suite including evaluations, tracing, and A/B testing — giving developers what Microsoft describes as unprecedented visibility into agent behaviours and outcomes.

For teams who have struggled to understand why their agents occasionally produce wrong answers, or how to systematically improve them over time, this observability infrastructure is the foundation they have been missing.

1,800+ Models and the Model-as-a-Service Expansion

Azure AI Foundry’s model catalogue has continued to expand in 2026, now offering access to over 1,800 models. The Model-as-a-Service approach means enterprise teams can access frontier models from OpenAI (including GPT-5.5 Instant, made generally available in May 2026), Anthropic, Meta, Mistral, and others through a standard API, within Azure’s compliance and security envelope, without managing GPU infrastructure.  Microsoft Foundry & Azure Open AI is the related training course here. 

This matters particularly for UK organisations in regulated sectors: access to frontier AI models through a platform with existing compliance certifications (ISO 27001, SOC 2, GDPR, NHS Digital, and others) removes a significant procurement and risk management barrier.

Azure AI Inference SDK Migration

One important technical change that development teams need to be aware of: the Azure AI Inference beta SDK is deprecated and will be retired on 26 August 2026. Teams must migrate to the generally available OpenAI/v1 API with a stable OpenAI SDK. This is a breaking change for any existing applications using the beta SDK — teams should begin migration planning immediately if they have not already.

How Azure AI Foundry Fits With the Rest of the Microsoft AI Stack

A question that comes up constantly with clients is where Foundry sits relative to Copilot Studio and Semantic Kernel. The answer in 2026 is clearer than it was twelve months ago:

• Azure AI Foundry is the platform layer: where you host models, run the Agent Service, manage evaluations, and deploy AI applications at scale
• Semantic Kernel is the developer SDK: C# and Python code that connects to Foundry’s services and orchestrates agent behaviour in your application
• Copilot Studio is the no-code/low-code layer: building agents that live in Microsoft 365 and connect to Foundry’s models and knowledge sources

Most sophisticated enterprise AI implementations in 2026 use all three, with Foundry providing the infrastructure, Semantic Kernel providing the developer control layer for custom agents, and Copilot Studio providing the business-user-facing agent experiences.

What Teams Need to Learn

Azure AI Foundry has meaningful depth. Effective production use requires understanding the Foundry portal and project structure, model deployment options and cost management, the Azure AI Agent Service for managed agent hosting, observability and the evaluation pipeline, the Red Teaming Agent for safety testing, integration patterns with Azure Data Lake, Azure AI Search, and Microsoft Fabric, and the migration from the deprecated Inference SDK to the stable API. For teams building production systems, this is not knowledge that can be acquired from documentation alone.

Train Your Team with JBI Training

JBI Training delivers expert-led, instructor-led AI solutions with Microsoft Foundry & Azure Open AI in London, online, and on-site across the UK. All courses are practical and hands-on.

Explore all Microsoft AI courses: https://www.jbinternational.co.uk/courses/microsoft-ai

Tags: Azure AI Foundry training UK, Azure OpenAI training 2026, AI solutions Microsoft Foundry course London, Azure AI agent service training, enterprise AI training UK 2026, GPT-5.5 Azure training