Topic 1 of 6
The shared responsibility model
Moving workloads to the cloud doesn't hand all security to the provider. This model defines what Microsoft protects and what stays yours, and it changes depending on whether you use on-premises, IaaS, PaaS, or SaaS. It's the foundation for almost every cloud security question on the exam.
Contents
What is the shared responsibility model?
In your own datacenter, the organization owns everything: the building, servers, network, operating system, applications, and data. In the cloud, some of those tasks move to the provider. The shared responsibility model makes clear who answers for each layer, so security can be planned without gaps.
On-premises, your own team handles the physical security of the building, maintaining and patching hardware, configuring network devices, securing operating systems, deploying and updating applications, protecting identities and access, and safeguarding data. That full control also means carrying the entire security burden.
The cloud changes that split. The "higher" the service (from IaaS to PaaS to SaaS), the more layers the provider takes on. But there's one idea the exam keeps coming back to: responsibility is shared, accountability isn't. You never delegate accountability for your data, your identities, your endpoints, and how you use the technology.
On-premises, IaaS, PaaS, and SaaS
On-premises
The organization manages and secures the entire stack, from the building and hardware to the operating system, applications, and data.
Examples: Your own servers in a company server room.
IaaS
Microsoft secures the physical datacenter, network hardware, and hosts. You manage operating systems, applications, the network controls you configure, and data.
Examples: Azure Virtual Machines, disks, virtual networks.
PaaS
Microsoft also handles the operating system and runtime. You answer for the application code, its configuration, access controls, and data.
Examples: App Service, Functions, Azure SQL Database, Storage.
SaaS
Microsoft manages the whole application and its infrastructure. You control who uses the service, your data, and tenant-level configuration.
Examples: Microsoft 365, Dynamics 365.
Rule of thumb
Ask yourself "who can touch this layer?". If you can sign in to the operating system (a VM), you patch it. If you only deploy code (App Service), the operating system isn't yours. If you only use the application (Microsoft 365), you're left with data, access, devices, and configuration.
Who handles what in each model
Matrix based on current Microsoft documentation. As you move from on-premises to SaaS, the provider takes on more layers.
| Responsibility | On-premises | IaaS | PaaS | SaaS |
|---|---|---|---|---|
| Customer data | Customer | Customer | Customer | Customer |
| Configurations and settings | Customer | Customer | Customer | Customer |
| Identities and users | Customer | Customer | Customer | Customer |
| Client devices | Customer | Customer | Customer | Shared |
| Applications | Customer | Customer | Shared | Shared |
| Network controls | Customer | Customer | Shared | Microsoft |
| Operating system | Customer | Customer | Microsoft | Microsoft |
| Physical hosts | Customer | Microsoft | Microsoft | Microsoft |
| Physical network | Customer | Microsoft | Microsoft | Microsoft |
| Physical datacenter | Customer | Microsoft | Microsoft | Microsoft |
Notice the first three rows: they're "Customer" in every column. And the last three: as soon as there's a cloud, they belong to Microsoft.
What always belongs to the customer
Regardless of the service model, Microsoft teaches that the customer always keeps these responsibilities:
💾 Data
Classifying it by sensitivity, deciding how to protect and encrypt it, and meeting data governance requirements. The provider secures the infrastructure that stores it, but you decide what data you collect, who accesses it, how long it's kept, and how it's protected.
🪪 Identities and access
Accounts, authentication methods, and permissions. A compromised identity gives access to resources even when everything else is locked down. Requiring MFA and applying least privilege is one of the customer's most important tasks.
💻 Endpoints
Laptops, tablets, phones, and computers that connect to the cloud. An attacker who compromises an endpoint can access services as its legitimate user. Device management, updates, and threat detection belong to the customer.
⚙️ Configuration choices
How you configure services, resource permissions, and network policies in your tenant. A misconfigured storage account or an overly permissive access policy exposes data even when the platform is secure.
Why this matters on the exam
Questions often describe a service and ask who protects a layer. If the layer is data, accounts, devices, or tenant configuration, the answer is the customer, even for SaaS.
Shared responsibilities
Between what's always yours and what's always the provider's, some layers change owner depending on the model:
Applications
In IaaS they're entirely yours. In PaaS and SaaS, Microsoft manages part of the stack, but you answer for the application configuration, code security, and access controls.
Network controls
In IaaS you configure all network security (firewalls, segmentation). In PaaS, Microsoft provides a baseline and you configure the application's network controls. In SaaS, the network is Microsoft's.
Client devices
In SaaS, Microsoft may provide device management tools, but protecting endpoints and keeping them compliant is still yours.
Microsoft, for its part, answers for the physical security of datacenters, the physical network, physical hosts, and the hypervisor, and in PaaS and SaaS also for operating systems, runtimes, and middleware.
Cloud security advantages
The model doesn't just move tasks around: it also gives you capabilities few organizations could build and sustain on their own.
What the provider invests in for you
- Physical datacenter security
- Hardware reliability
- Compliance certifications
- Threat intelligence
- Large-scale security monitoring
Typical on-premises gaps
- Patches delayed by stretched teams
- Hardware past its vendor support window
- Systems without monitoring tools
In the cloud, the provider patches the infrastructure and protects the physical side of its part of the stack. That doesn't remove your obligations, but it lets you focus effort on data, identities, and configuration, where it has the most impact.
Shared responsibility for AI
Current SC-900 material extends the model to AI-enabled solutions. AI systems can be influenced by the data and instructions they receive and produce outputs that may drive important decisions, so they bring new responsibilities. An AI application is thought of in three layers:
AI platform
Infrastructure, the model, and platform security controls, such as content filters or built-in safety systems.
AI application
The app you build or deploy on the platform: configuration, data sources, connectors, and enabled plugins.
AI usage
How people use it: what data and prompts they enter, acceptable use policies, and oversight.
As with the classic cloud, the split depends on the service type: with SaaS AI (for example, Microsoft 365 Copilot) the provider takes on more; with PaaS AI (for example, Azure AI Foundry) you build and configure more.
Usually the provider
- Physical infrastructure and model hosting
- Platform security controls, depending on the service tier
- Secure operation of the service's compute
Always the customer
- What data it shares with or lets the AI read
- Who can use the AI application and what they can do
- Secure configuration: connectors, logs, retention
- Mitigating prompt injection
- Training users and defining usage policies
Prompt injection
An attacker embeds instructions in user input or external data to manipulate the AI's behavior or output. It's mitigated with input validation, access controls, and monitoring, and it's the customer's responsibility.
Exam scenarios
| Situation | Responsible | Why |
|---|---|---|
| An Azure VM has an unpatched operating system | Customer | In IaaS, the VM's operating system belongs to the customer. |
| A power outage affects an Azure datacenter | Microsoft | Physical datacenter security and operation belong to the provider in every cloud model. |
| A Microsoft 365 user shares a confidential file with an outsider | Customer | Data, access, and tenant configuration remain the customer's in SaaS. |
| An App Service app's runtime needs updating | Microsoft | In PaaS, the operating system and runtime belong to the provider. |
| A storage account was left publicly accessible by mistake | Customer | It's a configuration decision: the platform is secure, the configuration isn't. |
| An employee pastes customer data into an AI assistant | Customer | How AI is used and what data it's given is always the customer's responsibility. |
Common exam traps
❓ In SaaS, is data security Microsoft's responsibility?
No. Data, identities, and tenant configuration belong to the customer in every model, including SaaS. Microsoft protects the application and infrastructure; it doesn't decide who accesses your files.
❓ In PaaS, does the customer patch the operating system?
No. In PaaS the operating system and runtime are Microsoft's. In IaaS, by contrast, you do patch your VMs' operating system.
❓ Does moving to the cloud mean delegating all responsibility?
No. Infrastructure layers are delegated, but never responsibility for data, identities, endpoints, and how the technology is used.
❓ Is physical datacenter security shared in IaaS?
No. In any cloud model, the datacenter, physical network, and physical hosts belong to the provider. Only on-premises are they the customer's.
❓ Are network controls in SaaS the customer's?
No. In SaaS the network is Microsoft's. In PaaS it's shared and in IaaS it's the customer's.
❓ With Microsoft 365 Copilot, who decides what data the AI can read?
The customer. Even if the AI platform belongs to the provider, the data that's shared and the access to it always belong to the customer.
Did you get this topic?
Test what you just learned
A company uses Azure SQL Database (PaaS). Who is responsible for patching the database server's operating system?
Ready to test "Shared responsibility model"?
Original questions, each with a full explanation.
Start a mock examSign in to save your progress.