Azure security baseline for Azure OpenAI

This security baseline applies guidance from the Microsoft cloud security benchmark version 1.0 to Azure OpenAI. The Microsoft cloud security benchmark provides recommendations on how you can secure your cloud solutions on Azure. The content is grouped by the security controls defined by the Microsoft cloud security benchmark and the related guidance applicable to Azure OpenAI.

You can monitor this security baseline and its recommendations using Microsoft Defender for Cloud. Azure Policy definitions will be listed in the Regulatory Compliance section of the Microsoft Defender for Cloud portal page.

When a feature has relevant Azure Policy Definitions, they are listed in this baseline to help you measure compliance with the Microsoft cloud security benchmark controls and recommendations. Some recommendations may require a paid Microsoft Defender plan to enable certain security scenarios.

The security profile summarizes high-impact behaviors of Azure OpenAI, which may result in increased security considerations.

For more information, see the Microsoft cloud security benchmark: Network security.

Description: Service supports deployment into customer's private Virtual Network (VNet). Learn more.

Configuration Guidance: Set up network rules to restrict access to your Azure AI services account. By default, enabling firewall rules blocks all incoming requests unless they originate from an allowed Azure Virtual Network subnet or a specified list of IP addresses. Authorization is required using Microsoft Entra ID credentials or an API key. You should first deny all traffic by default and then create rules that permit access from specific networks, ensuring a secure boundary for your applications.

Reference: Configure Azure AI services virtual networks

Description: Service native IP filtering capability for filtering network traffic (not to be confused with NSG or Azure Firewall). Learn more.

Configuration Guidance: Deploy private endpoints for all Azure resources that support the Private Link feature, to establish a private access point for the resources.

Reference: Configure Azure AI services virtual networks

Description: Service supports disabling public network access either through using service-level IP ACL filtering rule (not NSG or Azure Firewall) or using a 'Disable Public Network Access' toggle switch. Learn more.

Configuration Guidance: Disable public network access either using the service-level IP ACL filtering rule or a toggling switch for public network access.

Reference: Configure Azure AI services virtual networks

For more information, see the Microsoft cloud security benchmark: Identity management.

Description: Service supports using Azure AD authentication for data plane access. Learn more.

Configuration Guidance: No additional configurations are required as this is enabled on a default deployment.

Reference: How to configure Azure OpenAI Service with managed identities

Description: Local authentications methods supported for data plane access, such as a local username and password. Learn more.

Feature notes: While you can authenticate against Azure AI services using a single-service or multi-service subscription key, or use those keys to authenticate with access tokens, these authentication methods fall short in more complex scenarios that require Azure role-based access control (Azure RBAC). Avoid the usage of local authentication methods or accounts, these should be disabled wherever possible. Instead use Azure AD to authenticate where possible.

Configuration Guidance: Restrict the use of local authentication methods for data plane access. Instead, use Azure Active Directory (Azure AD) as the default authentication method to control your data plane access.

Reference: Authenticate with an access token

Description: Data plane actions support authentication using managed identities. Learn more.

Configuration Guidance: Use Azure managed identities instead of service principals when possible, which can authenticate to Azure services and resources that support Azure Active Directory (Azure AD) authentication. Managed identity credentials are fully managed, rotated, and protected by the platform, avoiding hard-coded credentials in source code or configuration files.

Reference: How to configure Azure OpenAI Service with managed identities

Description: Data plane supports authentication using service principals. Learn more.

Configuration Guidance: There is no current Microsoft guidance for this feature configuration. Please review and determine if your organization wants to configure this security feature.

Reference: How to configure Azure OpenAI Service with managed identities

Description: Data plane access can be controlled using Azure AD Conditional Access Policies. Learn more.

Configuration Guidance: Define the applicable conditions and criteria for Azure Active Directory (Azure AD) conditional access in the workload. Consider common use cases such as blocking or granting access from specific locations, blocking risky sign-in behavior, or requiring organization-managed devices for specific applications.

Description: Data plane supports native use of Azure Key Vault for credential and secrets store. Learn more.

Feature notes: Azure OpenAI service does not store secrets (e.g., passwords, etc.).

Configuration Guidance: Ensure that secrets and credentials are stored in secure locations such as Azure Key Vault, instead of embedding them into code or configuration files.

Reference: Develop Azure AI services applications with Key Vault

For more information, see the Microsoft cloud security benchmark: Privileged access.

Description: Azure Role-Based Access Control (Azure RBAC) can be used to managed access to service's data plane actions. Learn more.

Configuration Guidance: Use Azure role-based access control (Azure RBAC) to manage Azure resource access through built-in role assignments. Azure RBAC roles can be assigned to users, groups, service principals, and managed identities.

Reference: How to configure Azure OpenAI Service with managed identities

Description: Customer Lockbox can be used for Microsoft support access. Learn more.

Configuration Guidance: In support scenarios where Microsoft needs to access your data, use Customer Lockbox to review, then approve or reject each of Microsoft's data access requests.

Reference: Customer Lockbox for Microsoft Azure

For more information, see the Microsoft cloud security benchmark: Data protection.

Description: Tools (such as Azure Purview or Azure Information Protection) can be used for data discovery and classification in the service. Learn more.

Configuration Guidance: This feature is not supported to secure this service.

Description: Service supports DLP solution to monitor sensitive data movement (in customer's content). Learn more.

Configuration Guidance: Azure OpenAI services data loss prevention capabilities allow customers to configure the list of outbound URLs their Azure OpenAI services resources are allowed to access. This creates another level of control for customers to prevent data loss.

Reference: Configure data loss prevention for Azure AI services

Description: Service supports data in-transit encryption for data plane. Learn more.

Configuration Guidance: No additional configurations are required as this is enabled on a default deployment.

Description: Data at-rest encryption using platform keys is supported, any customer content at rest is encrypted with these Microsoft managed keys. Learn more.

Configuration Guidance: No additional configurations are required as this is enabled on a default deployment.

Reference: Azure OpenAI Service encryption of data at rest

Description: Data at-rest encryption using customer-managed keys is supported for customer content stored by the service. Learn more.

Configuration Guidance: If required for regulatory compliance, define the use case and service scope where encryption using customer-managed keys are needed. Enable and implement data at rest encryption using customer-managed key for those services.

Reference: Azure OpenAI Service encryption of data at rest

Description: The service supports Azure Key Vault integration for any customer keys, secrets, or certificates. Learn more.

Configuration Guidance: Use Azure Key Vault to create and control the life cycle of your encryption keys, including key generation, distribution, and storage. Rotate and revoke your keys in Azure Key Vault and your service based on a defined schedule or when there is a key retirement or compromise. When there is a need to use customer-managed key (CMK) in the workload, service, or application level, ensure you follow the best practices for key management: Use a key hierarchy to generate a separate data encryption key (DEK) with your key encryption key (KEK) in your key vault. Ensure keys are registered with Azure Key Vault and referenced via key IDs from the service or application. If you need to bring your own key (BYOK) to the service (such as importing HSM-protected keys from your on-premises HSMs into Azure Key Vault), follow recommended guidelines to perform initial key generation and key transfer.

Reference: Azure OpenAI Service encryption of data at rest

For more information, see the Microsoft cloud security benchmark: Asset management.

Description: Service configurations can be monitored and enforced via Azure Policy. Learn more.

Feature notes: Please refer to the policies for Azure AI services.

Configuration Guidance: Use Microsoft Defender for Cloud to configure Azure Policy to audit and enforce configurations of your Azure resources. Use Azure Monitor to create alerts when there is a configuration deviation detected on the resources. Use Azure Policy [deny] and [deploy if not exists] effects to enforce secure configuration across Azure resources.

For more information, see the Microsoft cloud security benchmark: Logging and threat detection.

Description: Service produces resource logs that can provide enhanced service-specific metrics and logging. The customer can configure these resource logs and send them to their own data sink like a storage account or log analytics workspace. Learn more.

Configuration Guidance: Enable resource logs for the service. For example, Key Vault supports additional resource logs for actions that get a secret from a key vault or and Azure SQL has resource logs that track requests to a database. The content of resource logs varies by the Azure service and resource type.

Reference: Monitoring Azure OpenAI Service

• See the Microsoft cloud security benchmark overview
• Learn more about Azure security baselines