Explainer

The cloud shared responsibility model explained

Learn how the shared responsibility model splits security duties between you and your cloud provider, and where the gaps usually appear.

By the Cloudverg engineering team · · 5 min read

What it is

The shared responsibility model is the split of security duties between a cloud provider and you, the customer. The provider secures the cloud itself: the buildings, the servers, the network backbone, the virtualization layer that keeps tenants apart. You secure what you put into the cloud: your data, your identities, your configurations, your code.

It is not one fixed line. Each provider publishes its own version, and the split moves with the service type. On a bare virtual machine you own the operating system, the patching and the firewall. On a managed database the provider patches the engine, and you still decide who can reach it and which network it sits in.

The model is about control, not blame. Whoever can change a setting owns the outcome of that setting. If you can make a storage bucket readable by the whole internet, you own that decision. The provider can only hand you a safe default and a log line when you change it.

One useful habit: read it as a map of who can fix what. That framing survives contract changes, reorganizations and new services far better than a list of bullet points copied from a vendor page.

How it works

Think in layers, bottom to top. Facilities, power and physical access sit with the provider. Above that, servers, disks and the network fabric. Then the hypervisor or container runtime. Then the guest operating system. Then the runtime and middleware. Then your application. Then your data and your identities. The line sits somewhere in that stack, and it slides depending on what you bought.

A virtual machine is the easiest case. The provider runs the hardware and the hypervisor. You patch the OS, close ports, manage SSH keys, run the firewall and take backups. Swap that for object storage and the picture changes: the provider keeps the service running and durable, while you own bucket policy, encryption choice, lifecycle rules and who holds the keys. On a managed Kubernetes service the provider runs the control plane and you own node images, network policies, secrets and pod security. On a SaaS product the provider covers almost everything inside the app, but you still provision users, connect single sign-on, classify what you upload and remove leavers.

Identity is the layer that never moves. Providers give you the tools, IAM policies, roles, federated login, but no provider decides which of your engineers gets admin on production. The same goes for data classification. The provider has no idea which records are regulated, which are customer secrets and which are test fixtures.

Managed services pull the line upward. You stop patching an engine, and you gain a new set of settings to get right: public endpoints, snapshot sharing, cross-account replication, logging destinations. The work does not vanish. It changes shape.

Why it matters for a business

Customer security questionnaires and audits ask for this split in writing. If you cannot show which controls you operate and which the provider operates, a reviewer will assume the gap is unmanaged. The review stalls, the deal slips, and someone has to rebuild the answer under deadline pressure.

Incident response depends on it too. At two in the morning, the model tells you who to call and what to do while you wait. A hypervisor failure is the provider's problem. A leaked access key in a public repository is yours. Confusing the two wastes the first hour, which is the hour that matters.

It also changes how you spend money. Knowing what is yours stops you from buying duplicate tooling for something already included, and stops you from skipping a control you assumed came free with the platform. Provider compliance certifications cover the provider's side only. Running on a certified platform does not make your workload compliant.

The model shapes your team too. Someone has to own guest patching, key rotation and access reviews. If nobody is named, those tasks quietly stop happening.

Common mistakes

Assuming backups are included because the service is durable. Many managed services replicate data across zones but do not retain older versions. A delete command or a bad migration propagates to every replica within seconds. Versioning and a separate backup account are usually your job.

Assuming encryption settles access. Encryption at rest protects disks that walk out of a data center. It does nothing when a policy allows anonymous reads on a bucket that holds the same data. Encryption and authorization are different controls, and you own the second one.

Treating the model as a legal shield. It does not transfer regulatory duty. Your customers, your regulator and your insurer still look at you first when data leaks, and they will ask what you configured.

Two more traps. Reading the general shared responsibility page and stopping there: the per-service documentation is where the real split lives, especially for managed databases, message queues and analytics services. And carrying one mental model across a whole account. A single subscription can hold virtual machines, serverless functions and SaaS add-ons, each with a different line.

How to get started

Build a service inventory first. List every service running in every account and subscription, then write a short table for each one: control, provider, you, or both. Pull the answers from the provider's per-service documentation rather than the marketing page. Keep the table next to your architecture notes so it gets updated when someone adds a service.

Next, work the controls in order of risk. Identity comes first: unused roles, over-broad permissions, keys that never rotate, accounts without multi-factor authentication. Then public exposure, which means storage policies, database endpoints, management consoles and anything reachable from the internet by default. Then logging, alerting and backups, checked against a benchmark you can point to.

Make the split part of your architecture review. Any new service gets one question: which controls move to us now? Add the answer to the inventory and to onboarding material, so the next engineer inherits the map instead of rediscovering it.

Some teams do this in-house. Others bring in outside help for a first pass. Cloudverg's Cloud security review, $6,500/month, is one option among others: a read-only review of up to five cloud accounts covering identity, public exposure, encryption, logging and backups, with findings ranked by risk and a specific fix for each.

Frequently asked questions

Who is responsible for security in the cloud?

The provider secures the physical infrastructure and the service itself. You are responsible for your data, identities, configurations and code. The exact line depends on the service: on a virtual machine you patch the operating system, while on a managed database the provider does the engine patching and you control access and network exposure.

Does the shared responsibility model mean the provider covers a data breach?

No. The provider covers failures in its own layer. If a leak comes from a public bucket policy, an over-permissioned key or an unpatched guest system, that sits on your side. Check the per-service documentation before you assume a control is included.

How do I document the split for an auditor?

Write a table per service listing each control and who owns it, and cite the provider's documentation for the split. Keep it with your architecture notes and update it whenever you adopt a service. Most auditors want the mapping plus evidence that your side is configured and monitored.

Monthly plan

Want help with this? Cloud security review

$6,500/month

2 to 3 weeks · Starts within 10 business days

A read-only review of identity, network exposure, encryption and logging across your cloud accounts, with a prioritized fix list.

Renews monthly. Cancel any time.
What is included

Related guides

More cloud security guides   Security and compliance

Smallest way to start

Get a senior architect on your problem this week

Ninety minutes a month on the cloud, AI or security question that is blocking you, a pre-read of your material and a written summary within two business days.

$450/month

Renews monthly. Cancel any time.