AWS, Microsoft Azure, and Google Cloud all use a shared responsibility model, but the boundary changes with the service you deploy. This guide gives cloud and compliance teams a practical matrix for assigning security ownership, collecting audit evidence, and reviewing control changes on a recurring schedule.
Overview
Cloud compliance is not achieved simply because a provider maintains certifications or operates secure data centers. Providers generally secure the underlying infrastructure, while customers remain responsible for how they configure services, protect data, manage identities, and operate workloads. The exact division depends on whether you use infrastructure, platform, or software services.
For example, a team running virtual machines usually manages the guest operating system, installed software, network rules, and workload data. A team using a managed database may no longer patch the database engine, but it still controls access, retention, encryption settings, network exposure, and the data stored there. A provider-managed application can shift more operational responsibility to the provider while leaving the customer accountable for users, permissions, business configuration, and lawful data use.
The matrix below is a planning tool rather than a substitute for provider documentation, contractual terms, or a formal risk assessment. Treat each row as a question: who performs the control, who verifies it, and what evidence demonstrates that it worked?
| Control area | AWS | Microsoft Azure | Google Cloud | Typical customer responsibility |
|---|---|---|---|---|
| Physical security | Provider for cloud facilities and core hardware | Provider for facilities and underlying infrastructure | Provider for facilities and underlying infrastructure | Review independent assurance reports, applicable regions, and contractual commitments |
| Identity and access | Shared; provider supplies identity capabilities, customer configures users, roles, federation, and privileged access | Shared; provider supplies identity capabilities, customer governs directory settings, roles, and access reviews | Shared; provider supplies identity capabilities, customer manages principals, roles, federation, and privileged access | Define least privilege, MFA requirements, joiner-mover-leaver procedures, and review frequency |
| Logging and monitoring | Shared; provider generates platform logs and customer enables, routes, retains, and reviews relevant events | Shared; provider exposes service and platform telemetry, customer configures collection, alerts, retention, and review | Shared; provider exposes audit and service logs, customer selects coverage, destinations, retention, and response workflows | Establish required log sources, time synchronization, retention, alert ownership, and evidence of review |
| Encryption | Shared; provider protects infrastructure and offers key-management features, customer selects settings and manages keys where applicable | Shared; provider offers encryption capabilities, customer configures data protection and key access | Shared; provider provides encryption mechanisms, customer governs configuration, keys, and access | Classify data, document encryption requirements, protect keys, and test recovery from key loss scenarios |
| Vulnerability management | Provider patches its infrastructure; customer patches guest operating systems, containers, applications, and dependencies as applicable | Provider patches its infrastructure; customer manages workload and application vulnerabilities | Provider patches its infrastructure; customer manages workload images, software, and dependencies | Maintain asset inventories, scanning coverage, remediation targets, exceptions, and verification records |
| Backup and recovery | Shared; provider operates service resilience, customer enables backups and defines recovery requirements | Shared; provider operates platform resilience, customer configures backup, retention, recovery, and geographic requirements | Shared; provider operates platform resilience, customer designs and tests workload recovery | Set recovery point and recovery time objectives, protect backups, test restoration, and retain test evidence |
| Incident response | Shared; provider handles incidents affecting its infrastructure, customer investigates and responds to its account and workloads | Shared; provider handles platform incidents, customer operates detection, containment, notification, and lessons-learned processes | Shared; provider handles provider-side events, customer handles workload, identity, and data incidents | Define escalation paths, preserve evidence, document notification decisions, and run tabletop exercises |
| Configuration and network security | Provider secures the cloud platform; customer configures accounts, networks, security groups, services, and workload settings | Provider secures the platform; customer configures subscriptions, networks, policies, resources, and workload settings | Provider secures the platform; customer configures projects, networks, policies, resources, and workloads | Use baseline configurations, change control, drift detection, and periodic exposure reviews |
The labels are intentionally broad. A managed service can reduce patching effort without eliminating the need for secure configuration, access governance, data protection, or audit evidence. For a control-by-control mapping, connect the matrix to your chosen framework. Teams working across SOC 2, ISO 27001, and NIST CSF can use the framework mapping guide to align control language.
What to track
Maintain a responsibility register for every production account, subscription, project, and material cloud service. At minimum, record:
- Service and deployment model: Identify the provider, region, service tier, workload owner, and whether the service is infrastructure, platform, or software managed.
- Control owner: Name the team accountable for implementation, the person or function performing the control, and the reviewer who verifies it.
- Configuration state: Record required settings for identity, network access, encryption, logging, backups, vulnerability scanning, and data retention.
- Evidence location: Link to configuration snapshots, access reviews, tickets, scan results, backup tests, incident records, or provider assurance documentation.
- Exceptions: Document the risk, business justification, compensating control, approving authority, expiration date, and remediation plan.
- Change signals: Track new services, changed regions, altered data classifications, provider service changes, and findings from security reviews.
Evidence should demonstrate operation, not merely intent. A policy saying that privileged access is reviewed is weaker than a dated access review showing scope, reviewer, decisions, and follow-up actions. For audit preparation, compare this register with the SOC 2 audit readiness checklist and maintain a clear evidence owner for each control.
Cadence and checkpoints
A useful cadence combines event-driven reviews with scheduled checks:
- Per change: Reassess responsibility when adding a service, changing a deployment model, moving data to another region, modifying identity architecture, or accepting a new integration.
- Monthly: Review high-risk configuration findings, privileged access changes, unresolved vulnerabilities, logging failures, backup exceptions, and expiring risk acceptances.
- Quarterly: Reconcile the cloud asset inventory with the responsibility register; review access, network exposure, key ownership, vendor commitments, and incident-response contacts.
- Before an audit or assessment: Sample controls across each provider and service model. Confirm that evidence covers the requested period and that owners can explain exceptions.
- Annually: Revalidate the shared responsibility assumptions, provider assurance documents, internal policies, recovery tests, and framework mappings.
Automated continuous compliance monitoring can identify drift, but automation does not decide whether a control is appropriate for a specific workload. Pair technical checks with human review of data classification, legal requirements, business impact, and documented risk acceptance. The continuous compliance monitoring guide provides a useful structure for metrics, ownership, and escalation.
How to interpret changes
Not every finding means the provider failed to secure the cloud. First determine which layer changed. A provider infrastructure issue, a customer configuration error, an application defect, and an unclear contract obligation require different responses.
When a new service is introduced, ask four questions:
- What does the provider operate, and what must the customer configure?
- Which data, identities, and network paths will the service access?
- Which controls become inherited, and which existing controls become irrelevant or need redesign?
- What evidence will prove the control operated during the audit period?
Escalate a change when it expands administrative access, removes a logging source, changes encryption or key ownership, affects backup recoverability, introduces a new subprocessor or integration, or places regulated data in a different location. A control that appears “managed” should still have an owner responsible for confirming the provider’s commitments and the customer’s configuration.
When to revisit
Use this article as a living checkpoint rather than a one-time comparison. Revisit the matrix at least quarterly for active environments and complete a formal review at least annually. Perform an additional review after a major provider service change, architecture redesign, new regulatory scope, significant incident, audit finding, merger, acquisition, or change in data classification.
End each review with three practical outputs: an updated responsibility register, a prioritized list of control gaps and exceptions, and an evidence collection plan with named owners and due dates. Add the review to your cybersecurity compliance calendar. If cloud services process personal data, also connect the technical review to your privacy governance process and, where appropriate, a documented data protection impact assessment.
Finally, confirm that the matrix matches the actual environment—not the architecture diagram from last year. Compare it with current inventories, identity records, configuration findings, backup tests, and vendor terms. That simple reconciliation is what turns the shared responsibility model into an operational cloud security and compliance control.