SECURITY TESTING · PRACTICAL GUIDE
API tenant isolation testing for SaaS and customer portals
Published: 23 October 2026
A customer portal may serve many organisations through the same application and database. Each organisation is a tenant: its users should see only the records and actions their permissions allow. A successful login does not, by itself, prove that those boundaries work.
API tenant-isolation testing examines whether an authorised user from one tenant can reach another tenant's information or operations. For UK businesses running SaaS products, supplier portals or client dashboards, this is a useful question to include explicitly in a security-testing brief.
Define the boundary before testing it
Start with a written description of who belongs to each tenant and what access they should have. Include exceptions such as an accountant supporting several clients or a supplier user invited to a shared project. An apparent cross-tenant result may be intended sharing, or it may be a serious defect. The policy determines which.
Separate tenant membership from job role. “Administrator” could mean administrator of one customer's workspace rather than administrator of the whole platform. Also distinguish ownership of a record from permission to approve, export or delete it.
OWASP's authorisation guidance separates identity verification from permission checks and recommends validating permissions on every request. Hiding a screen or button does not replace enforcement by the service handling the action.
Understand what object-level authorisation means
An API often receives an identifier for a business object, such as an order, project or document. The service must establish whether the requesting user may perform the requested operation on that particular object.
OWASP describes failures at this boundary as broken object-level authorisation. Unpredictable identifiers can make guessing harder, but are not a substitute for permission checks. A user can learn an identifier through an otherwise legitimate workflow.
For a commissioning manager, the key question is simple: does the test examine access to other customers' records using designated accounts and synthetic data, rather than stopping at whether the API requires a valid login?
Prepare a small but meaningful test set
Create two or more clearly labelled test tenants. Give each its own users and records, plus a documented sharing case if the application supports one. Choose recognisable synthetic values so an unexpected response can be traced without exposing real customer information.
A useful permission worksheet records:
- The acting account and its tenant membership.
- The account's role and any delegated access.
- The record or workflow being accessed.
- The permitted action and expected result.
- The environment and API version under review.
For example, a fictional maintenance platform could contain a North Test workspace and a South Test workspace. A North Test scheduler may update North Test appointments but must not view South Test appointments. A shared contractor may have narrower access to an explicitly assigned job. These are illustrative rules, not customer findings.
Ask the product owner to approve the worksheet. Otherwise, the test may expose uncertainty about the product's intended behaviour without resolving whether a control has failed.
Follow the customer journey beyond one endpoint
Map where a record travels after creation. A portal might display it in a list, include it in search, attach it to an export and make it available to a background job. Ask which of these paths are included in the assessment.
OWASP's multi-tenant security guidance considers isolation across database access, caches, sessions and asynchronous processing. This matters because correct behaviour in the main screen does not establish that every supporting component applies the same boundary.
Bring a realistic workflow to the scoping discussion. If a customer regularly creates a report and downloads it later, include that whole journey. If the application allows users to switch organisations, describe when that happens and which records should remain accessible afterwards.
Agree safe evidence and meaningful outcomes
Testing should use written authorisation, designated accounts and an agreed scope. Establish how potentially destructive actions will be simulated or excluded. There is no need to retrieve a real customer's full dataset to demonstrate an access-control problem.
Ask for evidence that connects the account, intended rule, affected route and observed result. An error code alone may not explain whether an operation took place. The report should describe what was confirmed and what remained unverified.
Keep legitimate access in the test plan too. A change that denies every request might prevent the unwanted access while also breaking the service. The remediation owner needs to know both whether the forbidden action is blocked and whether the intended workflow still works.
Turn the result into a repeatable check
After a finding, agree where the underlying permission decision should be corrected and which related workflows may use the same control. Record the affected release and deployment rather than treating a merged code change as proof that production is corrected.
Retain the agreed permission worksheet as a development and regression-testing reference. Review it when roles, sharing rules, integrations or tenant-switching behaviour change. New product features can create new access relationships even when existing routes remain unchanged.
SabreShield's web application and API security testing can examine designated roles and tenant boundaries within an agreed scope. Discuss the customer journeys that matter most, then agree the evidence needed to assess them and the retesting required after fixes.