Explore our new Cybersecurity Terms Glossary, explained by our experts.
Cybersecurity Terms Glossary
Tenant isolation is what stops one customer of a shared system reaching another customer’s data. It is enforced in compute, storage, identity, and keys — and it fails one query at a time.

What Is Tenant Isolation?
Tenant isolation is the set of controls that prevents one customer of a shared system from reading, modifying, or affecting another customer’s data. Every multi-tenant SaaS platform depends on it, as does every cloud provider, and the promise it makes is absolute in a way few other security properties are: a single successful cross-tenant request is a breach affecting someone who did nothing wrong.
Isolation strategies sit on a spectrum between dedicated infrastructure and shared infrastructure with logical separation.
| Model | How it works | Trade-off |
|---|---|---|
| Silo | Separate database, account, or cluster per tenant | Strongest isolation, highest cost and operational effort |
| Pool | Shared infrastructure, separation enforced in code and policy | Efficient and scalable, entirely dependent on correct implementation |
| Bridge | Shared compute with per-tenant data stores, or silos for premium tenants only | Pragmatic middle ground, two code paths to maintain |
Most platforms begin pooled because it is cheap, then add siloed tiers for regulated or enterprise customers who ask for them. That is a reasonable path provided the pooled tier is engineered with the same seriousness, which is where many products fall short.
The dominant failure mode is simply broken access control: an endpoint that checks authentication but not ownership, so changing an identifier in the request returns somebody else’s record. Second to that is the query that was written without a tenant filter during a refactor, surfacing months later.
Platform-level examples show the same pattern at scale. In August 2021 researchers disclosed a flaw in the Azure Cosmos DB notebook feature that allowed cross-tenant access to other customers’ database credentials. In April 2022 a vulnerability chain in Azure Database for PostgreSQL Flexible Server permitted access to other customers’ databases through the replication mechanism. In 2022 a set of issues in IBM Cloud Databases allowed an attacker to reach internal build infrastructure shared between tenants. None required unusual skill; each depended on one boundary being thinner than it looked.
Isolation cannot be assessed from a single account. The effective method is to create two tenants under your control, perform an action in the first, capture the identifiers it produces, and then attempt every operation on those identifiers from the second — reads, writes, deletes, exports, webhook subscriptions, file downloads, and support functions. Repeat with different roles within each tenant, since a tenant administrator often has endpoints an ordinary user does not.
Automate the core of this in your test suite, so an isolation check runs on every pull request rather than once a year. Then test the paths automation struggles with: file processing, asynchronous jobs, search indexes, and anything involving caching. We include this two-account methodology in our web application security services precisely because it finds defects that single-account testing cannot.
The same thinking applies internally. Separating customer data by tenant while running development and production from one cloud account leaves an obvious path around the boundary, which is the argument for multi-account architecture. Where tenants are mapped to Kubernetes namespaces, remember that a namespace is not a security boundary on its own: without network policy, Pod Security Standards, and scoped service accounts, a compromised workload reaches its neighbours.
Enterprise procurement and security questionnaires now ask directly how tenants are separated, and regulated buyers increasingly ask for evidence rather than an assertion. Under UK GDPR, a cross-tenant leak is a personal data breach for which you are likely to be the processor, carrying an obligation to notify the affected controllers without undue delay — and they in turn must consider notifying the ICO. Being able to describe your isolation model clearly, and show how it is tested, shortens sales cycles as well as reducing risk.
It raises the floor considerably, because an unscoped query cannot return another tenant’s rows. It does not solve everything: the application still routes connections, and a flaw in that routing, in shared caching, or in a support interface crosses the boundary regardless. Silo models also carry real operational cost in migrations, backups, and monitoring, which is why most platforms reserve them for larger customers.
Not by itself. Namespaces scope names and provide a place to attach policy, but without NetworkPolicy, Pod Security Standards, resource quotas, and scoped service accounts, workloads in different namespaces can still reach each other and share a node kernel. For genuinely untrusted tenants, use separate clusters or separate cloud accounts, or run workloads on a sandboxed runtime.
Use two tenants you control. Perform actions in the first, record every identifier returned, then replay every request from the second tenant’s session and confirm each one is refused. Cover reads, writes, deletes, exports, file downloads, webhooks, and administrative functions, and repeat across user roles. Automate the core cases in your pipeline so regressions are caught at the pull request, not in production.
Describe the model honestly: which layers are shared, which are dedicated, how the tenant context is derived and enforced, how encryption keys are scoped, and how often isolation is tested independently. Vague assurances invite follow-up questions and erode trust. A short architecture note with a recent test summary usually satisfies a security questionnaire faster than a lengthy policy document.
You May Be Interested In These
We use a few cookies that are needed to make this website work, and they are always on. We would also like to set analytics cookies, to see which pages are useful; functional cookies, so that our live chat works; and advertising cookies, to measure our campaigns. We will not set any of those until you say yes, and you can change your mind at any time from the Cookie Settings link in the footer.
Choose the cookies you are happy for us to use. Strictly necessary cookies are always on; everything else is off until you turn it on.
These cookies keep the website working: they hold your security checks, your form submissions and the choice you make here. The site cannot run without them, so they are always on.
These let our live chat open and remember your conversation, and let the newsletter sign-up work. Switch them off and the chat window will not appear.
These tell us which pages are read and how people find us, so that we can improve the site. The reports are about groups of visitors, never about you personally.
These let us measure our advertising and show you our adverts on other websites, such as LinkedIn, Google and Meta. Switch them off and you will still see adverts, but they will be less relevant.