← All posts
SOC 2 for multi-tenant SaaSSOC 2 tenant isolation evidencemulti-tenant SaaS audit scope

SOC 2 for Multi-Tenant SaaS: Scope and Isolation Evidence

Map a multi-tenant SaaS service for SOC 2, document shared boundaries, and keep evidence that tenant isolation works across requests, jobs, storage, and support.

A multi-tenant SOC 2 scope follows customer data through shared services and records the boundary each path enforces.
A multi-tenant SOC 2 scope follows customer data through shared services and records the boundary each path enforces.

For SOC 2, a multi-tenant SaaS team should describe the service that customers use, including the shared paths that handle their data, then show how its controls keep one tenant’s actions and data within the intended boundary. A shared database is one path. API authorization, background jobs, caches, object storage, support access, and recovery paths can cross tenants too. Start with a map of those paths before collecting screenshots for a control list.

What changes for SOC 2 in a multi-tenant SaaS service?

The AICPA Trust Services Criteria provide criteria for examining controls over the system management describes. They do not prescribe one SaaS tenancy architecture. Management defines the service, its commitments, risks, and controls. The independent CPA firm agrees on the examination scope and evaluates whether evidence is sufficient.

In a shared service, a mistake in a common path can affect several customers. AWS’s SaaS Lens describes pool, silo, and bridge models: resources may be shared, dedicated, or mixed. Its pool-isolation guidance explains why shared infrastructure needs explicit isolation controls. These AWS patterns are engineering guidance, not extra SOC 2 criteria.

The right first question is: where does a customer identity or tenant context enter each path, and where is it checked before data or work is returned? The SOC 2 scope guide covers the wider system boundary. Use a data flow diagram to make shared and dedicated paths visible.

Map the tenant boundary one path at a time

Use one row for each material path through the service. Include internal paths that a customer cannot call directly, because a worker or support tool can still read or change customer records.

Path Boundary to describe Example source evidence
Login and API How the server ties the caller to an authorized tenant and resource Identity configuration, authorization tests, reviewed code change
Database and search Where tenant ownership is enforced for reads and writes Schema or policy, query tests, migration review
Cache and object storage How keys, object access, and signed links stay within scope Configuration, access tests, sample denied request
Queue and worker How an authenticated producer and consumer preserve tenant authority Message contract, worker authorization test, deployment record
Admin and support Who may cross tenant boundaries, for what reason, and with what record Role grants, access review, approved support action log
Backup and recovery Who may restore shared data and how a single-tenant recovery avoids exposing another tenant Restore runbook, isolated test result, reviewer sign-off

The OWASP Multi-Tenant Security Cheat Sheet lists common failures in request context, shared data stores, jobs, storage, and logging. Use it to challenge your design. Do not copy every example into a SOC 2 control without checking whether it applies to your architecture and commitments.

An opaque tenant ID is not an authorization decision. For example, a request that changes tenant_id=alpha to tenant_id=beta should not gain access to beta’s records just because both IDs are valid. A background message carrying tenant_id=beta needs an authorized producer and a consumer that checks the target operation. A support role that can cross tenants needs a separate, approved path and reviewable use record.

Build one test that supports a narrow claim

Start with two test tenants, A and B, each with distinct synthetic users, records, files, and queued work. Use the same roles, API routes, database role, and worker path that production uses where a safe test environment permits it. Record the software version and configuration tested.

  1. As an A user, request A’s resource and confirm the authorized result.
  2. As the same user, request B’s resource through the same route. Preserve the denial response and confirm that no B data appears in the body, logs visible to A, or generated file.
  3. Run a tenant-scoped job and try an unauthorized B target through the real producer and consumer path. Check the resulting data and error record.
  4. Test a support or admin path with its actual privileged role, approval condition, and audit record. A normal A user must not inherit that access.
  5. Review any failure, fix it, rerun the affected test, and keep both results.

This is an example test, not an AICPA test script. It supports a claim about the paths, roles, data classes, and version exercised. Repeat or expand tests when a new data store, job type, access path, or deployment model changes the boundary. Keep source test output and the reviewed change that established the control, then connect them to the control record. Redact tokens and customer content from retained output. Keep sensitive raw artifacts in an access-controlled source system and reference them from the GRC workspace; do not commit secrets or personal data that may need erasure to Git. The control-testing guide shows how to record an outcome, exception, owner, and retest without hiding a failed run.

Treat dedicated customers as part of the same service map

A dedicated database or full customer stack can reduce one kind of shared-data risk, but it does not by itself remove shared control planes. Identity, deployments, logging, monitoring, billing, support access, and backup operations may still serve many customers. If a shared API routes to dedicated customer databases, document the shared API boundary and the database boundary separately. AWS calls this mix of shared and dedicated components a bridge model.

Do not claim that a dedicated tenant is fully isolated unless the architecture and tests support that claim. Conversely, do not assume a pooled design is outside SOC 2. Describe the controls management actually operates, the commitments it made, and the residual risks it accepts. Agree on the report description and evidence approach with the CPA firm.

If your team runs a separate stack for each customer, use the single-tenant SaaS guide to account for every deployment and check controls across the fleet.

Keep evidence current through the report period

For a Type 2 examination, a one-day demonstration does not describe operation over the whole period. Keep dated evidence of the relevant access reviews, change reviews, test runs, incidents, and exceptions for the controls in scope. When code changes a tenant boundary, connect the reviewed change, test result, deployment, and any follow-up. A failed test is a finding to resolve, not a result to delete.

FileGRC is a Git-native GRC workspace for SOC 2 work. JSON holds structured records, Markdown holds long-form work, and Git supplies the change history. An engineer can keep the scope narrative and boundary map with control, evidence, and finding records in the FileGRC workspace, then use an agent to locate missing owners or prepare a sourced auditor response. Agents can help with program work when they have access to current records, but they cannot turn starter proposals into proof that a tenant boundary worked. The application and source systems enforce the boundary, management operates the controls and reviews responses, and the independent CPA firm performs the examination and judges evidence sufficiency.

Open source · MIT

Run your SOC 2 program as files in Git.

Keep policies, controls, work, and evidence indexes in a repository your team and agents can inspect. Add optional FileGRC Autopilot email and Slack reminders to keep work moving.

Frequently asked questions

Does SOC 2 require a separate database for every SaaS customer?

No universal SOC 2 rule requires one database per customer. Management must describe the actual system, commitments, risks, and controls. A shared database needs an enforceable tenant boundary and evidence that the relevant access paths respect it. An independent CPA firm evaluates the controls and evidence for the scoped examination.

Is a passing tenant-isolation test enough for SOC 2?

No. A passing test supports a defined isolation claim for the paths and version tested. Management also needs to operate its other scoped controls and retain evidence over the relevant period. The CPA firm decides what testing and evidence are sufficient for the engagement.

Should shared workers and queues be in a multi-tenant SOC 2 scope?

If shared workers or queues process in-scope customer work or data, describe them and their tenant authorization path in the system boundary. Test that a job cannot act on another tenant's resources just because its message carries a tenant ID. Include the source configuration, tests, and operating records relevant to the control.

Does a dedicated customer instance remove tenant-isolation work?

A dedicated instance changes the isolation design, but shared identity, deployment, support, logging, and control planes may still affect customers. Document which components are dedicated and which are shared, then test the boundaries that exist in the real service.

Can FileGRC enforce tenant isolation in our SaaS app?

No. FileGRC is a Git-native GRC workspace for SOC 2 work. It keeps structured records in JSON and long-form work in Markdown with Git change history. Your application and source systems enforce tenant boundaries and produce evidence; your team operates the controls; the independent CPA firm performs the examination.