Cosmos DB Stores Your Most Valuable Data
Azure Cosmos DB is a globally distributed, multi-model database that often holds mission-critical application data. Its default configuration provides convenience for development — primary keys with full read-write access, public network access enabled, and no firewall rules. Hardening Cosmos DB means restricting network access, eliminating key-based authentication in favor of Azure RBAC, encrypting data with customer-managed keys, and enabling continuous monitoring for anomalous access patterns.
Threat Landscape and Attack Surface
Hardening Azure Cosmos DB requires understanding the threat landscape specific to this service. Azure services are attractive targets because they often store, process, or transmit sensitive data and provide control-plane access to cloud infrastructure. Attackers probe for misconfigured services using automated scanners that continuously sweep Azure IP ranges for exposed endpoints, weak authentication, and default configurations.
The attack surface for Azure Cosmos DB includes several dimensions. The network perimeter determines who can reach the service endpoints. The identity and access layer controls what authenticated principals can do. The data plane governs how data is protected at rest and in transit. The management plane controls who can modify the service configuration itself. A comprehensive hardening strategy addresses all four dimensions because a weakness in any single layer can be exploited to bypass the controls in other layers.
Microsoft’s shared responsibility model means that while Azure secures the physical infrastructure, network fabric, and hypervisor, you are responsible for configuring the service securely. Default configurations prioritize ease of setup over security. Every Azure service ships with settings that must be tightened for production use, and this guide walks through the critical configurations that should be changed from their defaults.
The MITRE ATT&CK framework for cloud environments provides a structured taxonomy of attack techniques that adversaries use against Azure services. Common techniques relevant to Azure Cosmos DB include initial access through exposed credentials or misconfigured endpoints, lateral movement through overly permissive RBAC assignments, and data exfiltration through unmonitored data plane operations. Each hardening control in this guide maps to one or more of these attack techniques.
Compliance and Regulatory Context
Security hardening is not just a technical exercise. It is a compliance requirement for virtually every regulatory framework that applies to cloud workloads. SOC 2 Type II requires evidence of security controls for cloud services. PCI DSS mandates network segmentation and encryption for payment data. HIPAA requires access controls and audit logging for health information. ISO 27001 demands a systematic approach to information security management. FedRAMP requires specific configurations for government workloads.
Azure Policy and Microsoft Defender for Cloud provide built-in compliance assessments against these frameworks. After applying the hardening configurations in this guide, run a compliance scan to verify your security posture against your applicable regulatory standards. Address any remaining findings to achieve and maintain compliance. Export compliance reports on a scheduled basis to satisfy audit requirements and demonstrate continuous adherence.
The Microsoft cloud security benchmark provides a comprehensive set of security controls mapped to common regulatory frameworks. Use this benchmark as a checklist to verify that your hardening effort covers all required areas. Each control includes Azure-specific implementation guidance and links to the relevant Azure service documentation.
Network Security
Private Endpoints
# Create private endpoint for Cosmos DB
az network private-endpoint create \
--name pe-cosmos-prod \
--resource-group rg-networking \
--vnet-name vnet-hub --subnet snet-private-endpoints \
--private-connection-resource-id "/subscriptions/{subId}/resourceGroups/rg-data/providers/Microsoft.DocumentDB/databaseAccounts/cosmos-prod" \
--group-id Sql --connection-name cosmos-pe-connection
# Disable public network access
az cosmosdb update --name cosmos-prod --resource-group rg-data \
--public-network-access DISABLED
IP Firewall Rules
# Allow specific IPs only (when private endpoints aren't used)
az cosmosdb update --name cosmos-prod --resource-group rg-data \
--ip-range-filter "10.0.0.0/24,203.0.113.10"
Virtual Network Service Endpoints
# Allow access from specific VNet subnet
az cosmosdb network-rule add --name cosmos-prod --resource-group rg-data \
--vnet-name vnet-hub --subnet snet-applications
Authentication and Access Control
Azure RBAC Instead of Primary Keys
Cosmos DB primary keys grant full read-write access to all data — they are the equivalent of root credentials. Migrate to Azure RBAC for data plane access:
# Disable key-based authentication (enforce RBAC only)
az cosmosdb update --name cosmos-prod --resource-group rg-data \
--disable-key-based-metadata-write-access true
# Assign data plane roles
az cosmosdb sql role assignment create \
--account-name cosmos-prod --resource-group rg-data \
--role-definition-name "Cosmos DB Built-in Data Reader" \
--principal-id "app-managed-identity-object-id" \
--scope "/dbs/appdb"
# For read-write access
az cosmosdb sql role assignment create \
--account-name cosmos-prod --resource-group rg-data \
--role-definition-name "Cosmos DB Built-in Data Contributor" \
--principal-id "api-managed-identity-object-id" \
--scope "/dbs/appdb/colls/orders"
RBAC assignments can be scoped to the account, database, or container level. Assign the narrowest scope that meets the application’s needs. The Cosmos DB Built-in Data Reader role provides read-only access; Cosmos DB Built-in Data Contributor provides read-write.
Disable Local Authentication
# Fully disable key-based auth (requires all apps to use AAD/RBAC)
az cosmosdb update --name cosmos-prod --resource-group rg-data \
--enable-analytical-storage false \
--disable-key-based-metadata-write-access true
Encryption
Customer-Managed Keys (CMK)
Cosmos DB encrypts data at rest by default with Microsoft-managed keys. For regulatory requirements, use customer-managed keys stored in Azure Key Vault:
# Create Cosmos DB with CMK
az cosmosdb create --name cosmos-prod --resource-group rg-data \
--key-uri "https://kv-prod.vault.azure.net/keys/cosmos-cmk" \
--assign-identity "[system]" \
--default-identity "SystemAssignedIdentity"
Always Encrypted (Client-Side Encryption)
For sensitive fields that should never be visible to the database engine, use Cosmos DB’s Always Encrypted feature. Data is encrypted in the client SDK before being sent to Cosmos DB and decrypted only in the client application. Even database administrators cannot read the encrypted fields.
Identity and Access Management Deep Dive
Identity is the primary security perimeter in cloud environments. For Azure Cosmos DB, implement a robust identity and access management strategy that follows the principle of least privilege.
Managed Identities: Use system-assigned or user-assigned managed identities for service-to-service authentication. Managed identities eliminate the need for stored credentials (connection strings, API keys, or service principal secrets) that can be leaked, stolen, or forgotten in configuration files. Azure automatically rotates the underlying certificates, removing the operational burden of credential rotation.
Custom RBAC Roles: When built-in roles grant more permissions than required, create custom roles that include only the specific actions needed. For example, if a monitoring service only needs to read metrics and logs from Azure Cosmos DB, create a custom role with only the Microsoft.Insights/metrics/read and Microsoft.Insights/logs/read actions rather than assigning the broader Reader or Contributor roles.
Conditional Access: For human administrators accessing Azure Cosmos DB through the portal or CLI, enforce Conditional Access policies that require multi-factor authentication, compliant devices, and approved locations. Set session lifetime limits so that administrative sessions expire after a reasonable period, forcing re-authentication.
Just-In-Time Access: Use Azure AD Privileged Identity Management (PIM) to provide time-limited, approval-required elevation for administrative actions. Instead of permanently assigning Contributor or Owner roles, require administrators to activate their role assignment for a specific duration with a business justification. This reduces the window of exposure if an administrator’s account is compromised.
Service Principal Hygiene: If managed identities cannot be used (for example, for external services or CI/CD pipelines), use certificate-based authentication for service principals rather than client secrets. Certificates are harder to accidentally expose than text secrets, and Azure Key Vault can automate their rotation. Set short expiration periods for any client secrets and monitor for secrets that are approaching expiration.
Advanced Threat Protection
# Enable Defender for Cosmos DB
az security pricing create --name CosmosDbs --tier Standard
Defender for Cosmos DB detects:
- SQL injection attempts in queries
- Anomalous access patterns (unusual geographies, unusual hours)
- Potential credential extraction (key listing operations from suspicious sources)
- Unusual data extraction volumes
Diagnostic Logging
az monitor diagnostic-settings create \
--name cosmos-diagnostics \
--resource "/subscriptions/{subId}/resourceGroups/rg-data/providers/Microsoft.DocumentDB/databaseAccounts/cosmos-prod" \
--workspace "/subscriptions/{subId}/resourceGroups/rg-monitoring/providers/Microsoft.OperationalInsights/workspaces/law-prod" \
--logs '[{"category":"DataPlaneRequests","enabled":true},{"category":"QueryRuntimeStatistics","enabled":true},{"category":"ControlPlaneRequests","enabled":true}]'
Enable DataPlaneRequests to audit all data access operations. Enable ControlPlaneRequests to track management plane changes (firewall rules, keys, RBAC). Use QueryRuntimeStatistics to identify expensive queries that might indicate abuse.
Additional Hardening
- Minimum TLS 1.2: Set via
az cosmosdb update --minimal-tls-version Tls12 - Disable cross-tenant replication: Prevent data replication to accounts in other tenants
- Configure backup: Use continuous backup for point-in-time restore; test restore procedures regularly
- Request unit limits: Set maximum RU/s limits at the database or container level to prevent cost-based denial of service
Defense in Depth Strategy
No single security control is sufficient. Apply a defense-in-depth strategy that layers multiple controls so that the failure of any single layer does not expose the service to attack. For Azure Cosmos DB, this means combining network isolation, identity verification, encryption, monitoring, and incident response capabilities.
At the network layer, restrict access to only the networks that legitimately need to reach the service. Use Private Endpoints to eliminate public internet exposure entirely. Where public access is required, use IP allowlists, service tags, and Web Application Firewall (WAF) rules to limit the attack surface. Configure network security groups (NSGs) with deny-by-default rules and explicit allow rules only for required traffic flows.
At the identity layer, enforce least-privilege access using Azure RBAC with custom roles when built-in roles are too broad. Use Managed Identities for service-to-service authentication to eliminate stored credentials. Enable Conditional Access policies to require multi-factor authentication and compliant devices for administrative access.
At the data layer, enable encryption at rest using customer-managed keys (CMK) in Azure Key Vault when the default Microsoft-managed keys do not meet your compliance requirements. Enforce TLS 1.2 or higher for data in transit. Enable purge protection on any service that supports soft delete to prevent malicious or accidental data destruction.
At the monitoring layer, enable diagnostic logging and route logs to a centralized Log Analytics workspace. Configure Microsoft Sentinel analytics rules to detect suspicious access patterns, privilege escalation attempts, and data exfiltration indicators. Set up automated response playbooks that can isolate compromised resources without human intervention during off-hours.
Continuous Security Assessment
Security hardening is not a one-time activity. Azure services evolve continuously, introducing new features, deprecating old configurations, and changing default behaviors. Schedule quarterly security reviews to reassess your hardening posture against the latest Microsoft security baselines.
Use Microsoft Defender for Cloud’s Secure Score as a quantitative measure of your security posture. Track your score over time and investigate any score decreases, which may indicate configuration drift or new recommendations from updated security baselines. Set a target Secure Score and hold teams accountable for maintaining it.
Subscribe to Azure update announcements and security advisories to stay informed about changes that affect your security controls. When Microsoft introduces a new security feature or changes a default behavior, assess the impact on your environment and update your hardening configuration accordingly. Automate this assessment where possible using Azure Policy to continuously evaluate your resources against your security standards.
Conduct periodic penetration testing against your Azure environment. Azure’s penetration testing rules of engagement allow testing without prior notification to Microsoft for most services. Engage a qualified security testing firm to assess your Azure Cosmos DB deployment using the same techniques that real attackers would employ. The findings from these tests often reveal gaps that automated compliance scans miss.
Hardening Checklist
- Network: Private endpoints; disable public access; IP firewall for any remaining access
- Authentication: Azure RBAC for data plane; disable primary key authentication
- Encryption: CMK for at-rest encryption; Always Encrypted for sensitive fields
- TLS: Minimum TLS 1.2
- Monitoring: Diagnostic logs to Log Analytics; Defender for Cosmos DB
- Access scope: Narrowest RBAC scope per application (container-level when possible)
- Backup: Continuous backup with regular restore testing
For more details, refer to the official documentation: Azure Cosmos DB overview, Best practices for Azure Cosmos DB .NET SDK, Diagnose and troubleshoot Azure Cosmos DB request timeout exceptions.