Home / Technological controls

What does ISO 27001 A.8.27 require for secure architecture?

ISO 27001 A.8.27 concerns the architecture and engineering principles used to keep systems secure through design, development and change. In practice, an architecture review needs to explain how the system protects information. Later checks establish whether those protections still work as new services connect and permissions change.

Control A.8.27

Updated November 4, 2026

A new integration can bypass a boundary that passed its architecture review. Reports arrive on time and the application stays available, while the integration uses a connection nobody assessed. The next review needs to account for that connection and the access it permits.

Architecture commitments

A.8.27 connects security principles to the choices made when systems are built or changed. The architecture explains where those principles apply across applications, infrastructure and external services. Engineers need enough detail to understand the reasons for a design before they modify it.

A decision record can supply that detail: the information being protected, the permitted way to handle it and the assumptions behind the choice. With those reasons available, a reviewer can identify which part of a proposed change needs closer examination.

Engineering principles

Engineering principles help a reviewer judge a proposed design. Least privilege concerns the authority a component needs for its work. Layers of protection reduce reliance on a single safeguard. Failure behavior matters because a dependency can stop working while sensitive information still needs protection.

NIST's engineering guidance discusses these principles. Their value in a review comes from applying them to a specific choice. For example, least privilege can determine whether a reporting service needs to read selected records or can also modify the database.

Trust boundaries

At a trust boundary, the system checks whether a request is allowed to proceed. External services often cross such a boundary when they contact an application. Components inside one environment can also have different permissions, requiring checks between them as information moves.

Each connection on the diagram should have an identifiable purpose. The reviewer needs to know which identity uses it, what that identity can request and where access is enforced. Those details make it possible to compare the diagram with actual permissions and connections.

Changed assumptions

Architecture drift occurs when the running system departs from a security assumption behind its design. A new integration might connect directly to a database that external services previously reached only through the application. The new path changes where access decisions take place.

Routine service checks may continue to pass because the integration returns the expected reports. Availability therefore leaves an important question unanswered: does the new connection preserve the intended protection? Comparing current information flows with the approved design gives the reviewer a way to investigate.

Design review

Changes to trust relationships deserve a design review. An external dependency, broader service permissions or another route to sensitive information can change the reasoning behind an earlier approval. Routine changes can follow established patterns where the assumptions supporting those patterns remain valid.

Systems security engineering addresses security throughout a system's life. For the reviewer, that means considering what a change does to existing protections. Identifying the affected information and the consequences of a failed protection helps determine how much review the change requires.

Design trade-offs

A correction needs to protect information and preserve the operations the service requires. Returning an integration to the approved application route may achieve both. Where requirements have changed, the owner may need to consider a replacement design and assess the access it introduces.

The choice depends on the work the integration performs. Reducing a permission may be enough when access is too broad. A separate interface may be appropriate when the required operation needs its own checks. The review record should explain why the selected approach fits.

Approved exceptions

An accepted exception needs to remain understandable after the person who requested it moves on. Its purpose, owner and review conditions explain why the departure was allowed. Without that context, a temporary connection can remain in use long after its original purpose ends.

The approval also depends on any safeguards agreed at the time. A different identity or a changed network route may invalidate those safeguards. Recording these dependencies gives the owner a reason to review the exception when circumstances change, even if the connection still works.

Checks between reviews

Permissions, network rules and deployment configuration can show whether an architecture decision still holds. Recurring checks compare these settings with an agreed state. Each check needs to identify which protection it assesses, so a failed result has a clear meaning for the reviewer.

ZeroTB performs these checks against approved commitments in connected systems. Its observations help the architecture owner investigate a change and decide what access remains acceptable. The service workflow connects that decision to an approved correction and a check of the resulting state.

Control restoration

Restoring a boundary requires attention to how the unwanted connection was created. If it came from deployment configuration, the next deployment could recreate it after a manual repair. The correction needs to cover that configuration as well as the current connection.

The owner first decides what access the service should retain and approves the material change. The implementation and design record can then be updated together. If the correction introduces another route, testing it before removing the existing route helps preserve the required service.

Verification tests

A successful correction has two results: the prohibited path is closed, and the approved operation still works. An integration test can check both by attempting the direct database connection and requesting a report through the permitted interface. Each attempt tests a different expectation.

Useful test evidence identifies the environment, the service identity, the attempted operation and when the test ran. A failed result shows what still needs attention. If the system cannot be checked, its state remains unknown and the gap stays open until verification is possible.

Evidence for later changes

A later reviewer should be able to understand why a correction was accepted. The Verified Control Record connects the commitment, observed gap, approved action and checked result. Keeping them together preserves the reasoning that could otherwise disappear between a design discussion and a completed ticket.

Some of this evidence may also help with technology controls or security risk reviews under another framework. Reuse depends on what that requirement covers and which systems are in scope. The following mapping identifies related requirements so a reviewer can make that assessment.

Audit evidence

A reviewer needs to trace the design's security purpose through the change decision to the behavior checked afterward.

RequirementWhat an auditor asks forWhere it comes from
Defined principlesThe principles used in the design and the security purpose of the selected protectionsEngineering guidance and design decisions showing how the principles apply to the system
Understood boundariesThe information allowed across each boundary, the requesting identities and the checks applied thereCurrent diagrams compared with deployed connections, access settings and the configuration that controls them
Reviewed changesThe assumption affected by a change and the owner's reason for accepting the designReview records and approvals identifying the dependency, interface or deployment affected by the change
Bounded exceptionsThe purpose of a departure, its responsible owner and the safeguards required to retain itException records with review conditions and the dependencies on which the agreed safeguards rely
Verified restorationThe correction approved for a gap and results for both prohibited and permitted operationsDated tests and system observations attached to the approved change and its control record
FrameworkControlWhat it covers
ISO 27001A.8.27Security principles used in system architecture and engineering, including decisions about later system changes
SOC 2CC5.2General controls over technology; a design decision can explain how a selected protection supports an objective
HIPAA164.308(a)(1)Security management for electronic protected health information, including analysis and management of risks to that information

Example of restoring an application boundary

This example follows a reporting integration that bypassed the application's access checks by connecting directly to its database. The correction returns reports to the approved application route. Testing must establish that reporting still works and that the direct connection is blocked.

Restore the approved reporting route
An application boundary record: a reporting service connecting straight to the database, the owner's approved return to the application route, a report completing through the API while direct access is refused, and the decision kept with the results.
  1. Bypassed application boundary. The new database connection lets the reporting service avoid checks applied by the application.

  2. Approved route change. The owner approves the application route and removal of the service's direct database permission.

  3. Verified reporting access. A report completes through the application, while a test of direct database access fails.

  4. Retained design decision. The approval and dated test results explain why the reviewer accepted the restored boundary.

Architecture review checklist

Identify the protection

Select an architecture decision and establish which information, operations and system boundaries its protection covers.

Inspect current connections

Compare the approved information flow with deployed permissions and connections, including dependencies added since the design review.

Decide the intended design

Assess the changed assumption and approve a correction or revised design that preserves the required protection.

Test both outcomes

Verify that prohibited requests fail and required operations still succeed through the approved route after the change.

Preserve the rationale

Retain the design decision, approval and dated results with the owner and conditions for the next review.

Start with the Pilot

Talk with our team about keeping architecture controls effective as systems change. The eight-week Pilot covers up to 25 commitments and two real control gaps, with approved corrections, verified results and the records that explain them.