
TL;DR
Persistent compromise changes the design objective for enterprise infrastructure. Protecting workloads against ordinary failures is not sufficient when an attacker may retain access through a privileged account, management appliance, automation identity, remote-access service, or altered configuration. VMware Cloud Foundation, Azure Local, and Nutanix provide different management and enforcement capabilities, but none eliminates the need to separate production authority from the systems used to investigate, contain, and recover it.
VMware Cloud Foundation 9.1 provides distinct management, control, and workload layers through components such as vCenter, NSX, VCF Operations, and VCF Management Services. Connected Azure Local relies on Azure and Azure Arc for supported cloud-managed operations while retaining local infrastructure dependencies. Nutanix operates through Prism Central, cluster-level Prism Element, AHV, and Controller VMs, with Flow Network Security available for appropriate workload-segmentation designs. Their security mechanisms are not feature-equivalent, and their behavior under compromise depends on how the enterprise configures and operates them.
Design the platform so that compromise of one administrative path does not automatically grant control over every workload, security policy, backup, and recovery action. The required evidence is practical: which identities can act, which boundaries they can cross, what continues operating during containment, and how trusted recovery can proceed without depending entirely on the compromised environment.
In This Article
- Introduction
- Define the Failure Scenario Before Selecting Controls
- The Five Boundaries That Determine Whether Compromise Can Persist
- VMware Cloud Foundation: Protect Management Before Workloads
- Azure Local: Separate Azure Authority From Local Infrastructure Control
- Nutanix: Preserve Trust Across Prism Central, Prism Element, and AHV
- Compare Responsibilities, Not Product Names
- Design a Control Boundary That Can Survive Compromised Management
- Make Recovery Copies Useful After a Compromise
- A Phased Implementation Path
- Test Containment Without Using a Live Incident as the First Exercise
- External References
Introduction
Imagine an enterprise running three infrastructure environments.
A VMware Cloud Foundation deployment hosts core business applications. Azure Local supports distributed infrastructure across several operational sites. Nutanix runs a separate portfolio of virtual machines and application services.
Each environment has healthy clusters, completed backup jobs, documented recovery objectives, and security monitoring.
An attacker compromises a privileged automation credential used to administer one of the environments.
The attacker does not immediately disrupt production. Instead, the credential is used to examine management configuration, create additional access, and investigate backup and recovery systems.
Several weeks later, the original credential is revoked.
The security team now faces a more difficult question: what other authority could have been created while the credential was valid?
The answer may involve local administrative accounts, service identities, API credentials, altered network policies, unauthorized virtual machines, modified automation, or persistent code on management appliances.
The initial credential was the entry path. The continuing risk is the authority that may have survived it.
The dormant-access guide details the persistence paths that can survive credential revocation.
This is not a theoretical concern. Mandiant’s September 2025 BRICKSTORM investigation documented long-term access to infrastructure appliances and VMware management systems. Its findings included attacker-created local accounts, modified startup mechanisms, and use of vCenter to clone sensitive virtual machines. Mandiant reported an average undetected dwell time of 393 days across its investigated intrusions. See Mandiant’s BRICKSTORM investigation.
That historical figure is specific to the BRICKSTORM investigation. It is not a general dwell-time benchmark for VCF, Azure Local, or Nutanix, and the investigation did not establish AI-controlled malware.
It does, however, demonstrate why infrastructure management must be part of the enterprise threat model.
This article takes the next architectural step in the series.
Start with the AI cyberattack series overview for the evidence and boundaries behind these design decisions.
Earlier installments examined AI-assisted attacker frameworks, persistence, supplier trust, exposed systems, and compromised development tooling. Here, the objective is to translate those findings into a platform-neutral containment architecture with platform-specific implementation boundaries.
The scope is pre-compromise architecture, operational controls, and recovery readiness, using VMware Cloud Foundation 9.1-family documentation, Microsoft’s connected Azure Local architecture guidance, and current Nutanix security guidance. The exact installed release, entitlement, hardware compatibility, and security configuration must still be established for each environment. See NIST’s cyber-resilient systems framework.
Cyber recovery after management-plane compromise is a separate problem. This article defines the design conditions that make that recovery credible.
Define the Failure Scenario Before Selecting Controls
A design for persistent compromise should begin with an explicit failure assumption.
For this article, the primary scenario is that an attacker has already gained meaningful access to a production infrastructure environment. The attacker may have created additional identities, changed management configuration, installed unauthorized software, or attempted to weaken detection and recovery.
The design must preserve the enterprise’s ability to respond even when some previously trusted systems can no longer be assumed trustworthy.
That is more demanding than designing for a host failure.
A failed host might be replaced by healthy infrastructure. A compromised management service may still function normally while executing unauthorized instructions.
A successful API response therefore does not establish a trustworthy control plane.
Separate Three Classes of Compromise
Workload compromise begins inside a virtual machine, container, application, or ordinary operating-system identity. The architectural objective is to prevent the attacker from using that position to reach infrastructure management or unrelated workloads.
Privileged identity compromise begins with access capable of changing infrastructure resources, permissions, network policy, or automation. The objective is to constrain the identity’s effective reach and preserve an independent means of restricting it.
Management-plane compromise occurs when the management component, its configuration, execution environment, or administrative authority can no longer be trusted. The response may require rebuilding management services, examining changes they made, and reestablishing the policies that control workloads.
These scenarios overlap, but they require different proof.
A microsegmentation rule may help contain a compromised VM. It is not sufficient if the attacker controls the system authorized to delete that rule.
Similarly, a backup may restore a VM while leaving the compromised administrative identity untouched.
The architecture should anticipate these different failure modes without assuming that every incident requires a complete platform rebuild.
The Five Boundaries That Determine Whether Compromise Can Persist
There are five boundaries worth testing across every enterprise infrastructure platform.
Management Access
Which users, services, networks, and tools can reach the infrastructure management interfaces?
The inventory must include normal administrator paths, automation services, remote-support connections, hardware management controllers, and emergency access.
A management interface that is isolated from application workloads may still be accessible through a shared jump host or privileged service identity.
Review actual reachability, not only the VLAN assignment.
Identity and Authority
Which identities can create or modify other identities, grant permissions, change certificates, or access secrets?
The central concern is authority expansion.
A compromised operator account is more consequential when it can create a second administrator, alter an identity-provider integration, or change the credentials protecting backup infrastructure.
The review should distinguish authentication from authorization and examine the effective permissions available after authentication succeeds.
Workload and Policy Enforcement
Which controls prevent one compromised workload from reaching another workload or management service?
Identify the actual enforcement location and the system that administers its policy.
Network segmentation is weaker when the same compromised administrator can change the policy without independent approval or monitoring.
For high-consequence boundaries, evaluate whether an additional independently administered control is required.
Evidence Integrity
Which records would establish unauthorized access, management changes, and attempted persistence?
Management logs are valuable, but they should not be the only evidence available when their source is compromised.
Forward important security and administrative events to a separately protected destination where supported.
Preserve the distinction between a centralized monitoring interface and independently controlled evidence storage.
Recovery Authority
Who can access backup copies, restore infrastructure, manage encryption keys, and authorize workloads to reconnect?
The recovery path should not depend exclusively on the identities or management services suspected of compromise.
That does not necessarily require duplicating the entire management platform. It requires a trusted administrative and evidence path appropriate to the failure scenario.
An Architecture for Independent Control
The following conceptual diagram illustrates these boundaries.

Independence is a design property that must be verified.
Two administrative accounts using the same identity provider, the same privileged workstation, and the same password vault may not be independent during a compound compromise.
Likewise, two backup copies administered by the same compromised credential may be vulnerable to the same unauthorized deletion or configuration changes.
The architecture should make these dependencies visible before an incident.
VMware Cloud Foundation: Protect Management Before Workloads
VMware Cloud Foundation 9.1 provides an integrated private-cloud operating model across infrastructure, networking, operations, automation, and workload services.
The relevant management components include vCenter Server, NSX Manager, SDDC Manager, VCF Operations, VCF Management Services, VCF Identity Broker, and associated administrative interfaces.
These components do different jobs.
VCF Operations supports fleet visibility and operational management. SDDC Manager and the associated lifecycle services coordinate platform lifecycle responsibilities. vCenter manages virtual infrastructure resources. NSX provides networking and security functions through separate management, control, and data-plane components.
An attacker does not need control of every component to produce serious consequences.
Management Network Isolation
Broadcom’s October 30, 2026 guidance recommends isolating infrastructure management from normal workload and user networks. See Broadcom’s management-isolation guidance.
That includes management appliances, ESX management interfaces, and the dependencies required to administer them.
The guidance distinguishes ESX management traffic from other management components and describes separate ESX management VLANs for individual workload domains.
A separate VLAN does not, by itself, establish security isolation.
The enterprise needs enforcement between the networks, with approved administrative paths and limited inbound and outbound connectivity.
For persistent-compromise planning, that boundary should include:
- Approved administrator and automation sources.
- Management appliance and ESX management destinations.
- DNS, time, certificate, and identity dependencies.
- Logging, backup, and recovery interfaces.
- Hardware management controllers.
- Required update and vendor-support connectivity.
A compromised management appliance should not have unrestricted outbound internet access merely because the management network is considered trusted.
Mandiant’s BRICKSTORM findings make that concern concrete. The investigated backdoors used outbound communications and targeted infrastructure appliances that were not covered by ordinary endpoint security tooling.
Broadcom’s management-isolation guidance similarly recommends restricting unnecessary egress and using supported proxy or offline software-depot workflows. See the VCF 9.1 hardening guidance.
Separate Workload Segmentation From Management Protection
VMware NSX can provide distributed firewalling and additional security capabilities through the applicable VCF and vDefend configuration.
Those controls can restrict east-west application traffic, establish workload isolation, and reduce opportunities for lateral movement.
However, NSX security features must be configured and operated. The presence of NSX does not mean that every workload is automatically microsegmented.
The enterprise should identify which protected groups, policy rules, and network paths enforce each critical boundary.
It must also examine who can change those policies.
If an attacker has obtained administrative control of the relevant NSX policy plane, the ability to alter segmentation becomes part of the incident.
A boundary protecting critical management access may therefore require enforcement that cannot be modified using the same compromised credentials.
The exact division between NSX, physical firewalls, management networks, and workload controls depends on the deployed topology.
Treat vCenter as a High-Impact Administrative System
Mandiant documented attackers using vCenter to clone sensitive virtual machines, including domain controllers and systems associated with identity or secrets.
A powered-off clone can contain valuable data even when guest security agents never execute.
The defensive review should therefore consider more than VM console access.
It should include the ability to clone, export, snapshot, reconfigure, move, or delete sensitive VMs, as well as access to their storage and backup copies.
Apply the least privilege supported by the business workflow, monitor sensitive operations, and prevent ordinary automation from inheriting broad administrative authority unnecessarily.
Review local accounts and service identities separately from federated access.
A change to an external identity provider should not silently invalidate every planned recovery path.
Preserve Management Recovery Dependencies
VCF 9.1 provides coordinated lifecycle and recovery-related capabilities, but infrastructure protection and trusted management recovery remain separate design tasks.
The recovery plan needs to establish how affected vCenter, NSX, management services, identity integrations, certificates, and supporting network paths will be restored or reestablished.
For management-plane compromise, a recent appliance backup may contain malicious configuration or unauthorized accounts.
That creates two questions:
- Can the management component be restored using a supported procedure?
- Can its resulting identity, configuration, and administrative authority be trusted?
The management-plane recovery guide addresses the deeper recovery sequence. The pre-incident design requirement is to protect the recovery sources, credentials, documentation, and network access necessary to make that decision.
Azure Local: Separate Azure Authority From Local Infrastructure Control
Azure Local introduces a different management relationship.
In connected deployments, Azure provides management and governance interfaces for supported Azure Local resources through Azure Resource Manager and Azure Arc.
The physical infrastructure, local operating system, clustering components, storage, networking, and workloads remain in customer-owned environments.
The platform therefore has both cloud and local security dependencies.
Microsoft’s Azure Local Well-Architected guidance explicitly distinguishes platform responsibilities from workload responsibilities and identifies identity, network, hardware management, connectivity, and recovery as separate design concerns. See Microsoft’s Azure Local architecture guidance.
Azure Arc Is a Management Dependency
Azure Arc resource bridge connects supported on-premises infrastructure with Azure-based management capabilities.
It enables supported resources to be represented and managed through Azure.
If the resource bridge becomes unavailable, Azure management operations that depend on it may fail while existing local virtual machines continue running.
That is an availability distinction.
It is not a guarantee that every operation remains available indefinitely without Azure connectivity.
It is also not a complete cyber-containment strategy.
A compromised Azure identity with sufficient control-plane authority and a compromised local administrative account represent different problems.
The incident plan must establish which actions are possible through each path and how to restrict them independently.
Do Not Treat Arc Resource Bridge as a Disposable Appliance
Microsoft’s maintenance documentation specifically warns against manually deleting or rebuilding the Azure Arc resource bridge for Azure Local without Microsoft guidance. See the resource-bridge maintenance guidance.
Unsupported removal can disrupt registrations, resource associations, and management operations.
For Azure Local, Microsoft’s general resource bridge maintenance guidance directs operators to Microsoft Support for major recovery operations.
An incident response playbook should not include generic instructions to delete the bridge and recreate it as if it were an ordinary workload VM.
The appropriate architecture is to identify the supported recovery path in advance, maintain the component’s lifecycle health, protect its dependencies, and retain the information required for an authorized recovery.
Azure Local Security Baselines Require Supported Management
Connected Azure Local hyperconverged deployments include security mechanisms such as application control, Credential Guard, BitLocker, and a managed security baseline.
Microsoft’s supported customization procedure depends on the installed release. In Azure Local 2610 and later, Set-AzsSecurity can customize individual baseline settings while drift control remains enabled, and cluster intent takes precedence over conflicting local values. Azure Local 2609 and earlier use a different procedure for protected settings. Use the release-specific guidance instead of applying a generic local override. See Microsoft’s security-baseline guidance.
The platform may reapply configured security intent during updates.
That is useful for consistency, but it creates an operational requirement: administrators should not use unsupported registry, Group Policy, or ad hoc configuration changes to implement security controls that the product manages through its own workflows.
The incident response team should distinguish legitimate baseline enforcement from unauthorized configuration.
It should also know which settings are managed centrally and how emergency changes can be safely authorized.
Azure Local Network Security Groups Have a Defined Scope
Azure Local supports network security groups for applicable Azure-managed virtual machines in supported software configurations.
Microsoft’s documentation specifies Azure Local 2506 or later with the required OS build family, enabled software-defined networking, and the supporting Azure Arc-managed resource model. See the Azure-managed NSG prerequisites and limitations.
The documented network security group capability applies to Azure Local VMs created and managed through supported Azure interfaces.
It should not be assumed to apply identically to every locally provisioned VM or to the separate on-premises software-defined networking operating model.
This matters during architecture review.
If the design contains a mixture of Azure-managed VMs, locally managed VMs, physical services, and legacy networking, the organization needs to identify where each rule is actually enforced.
A policy configured in Azure is not evidence that every workload network path is protected.
Connected and Disconnected Operations Are Different Architectures
Microsoft also documents a disconnected-operations model for Azure Local.
That is a distinct deployment and management architecture, not an emergency switch that automatically converts an ordinary connected environment into a disconnected one.
The current Azure Local Well-Architected guidance used here primarily addresses connected hyperconverged deployments.
An organization with a disconnected or sovereign requirement must validate that design against the separate Microsoft documentation and support conditions.
For persistent-compromise planning, the larger lesson is that local workload availability and trusted administrative control are different capabilities.
Both require evidence.
Nutanix: Preserve Trust Across Prism Central, Prism Element, and AHV
Nutanix has its own management and control boundaries.
Prism Central provides centralized infrastructure management, identity and authorization capabilities, and access to platform services.
Prism Element provides cluster-level administration.
AHV hosts and Controller VMs (CVMs) supply the underlying virtualization and storage operating environment.
Those components must be considered separately during a management-plane threat assessment.
Centralized Management Is Not the Entire Trust Boundary
Prism Central can provide centralized administration across clusters, but cluster-level configuration and access remain relevant.
An attacker with broad administrative permissions may be able to affect workloads, networking, or recovery-related settings.
The enterprise should identify the actual scope of each administrative role, including local accounts, directory integrations, and service accounts.
Nutanix’s public Prism Central RBAC guidance describes built-in and custom roles, authorization policies, and resource-scoped access for local users and directory users or groups. Verify the available identity types, service-account capabilities, and authorization behavior against the Security Guide for the installed release. See Nutanix’s Prism Central RBAC explanation.
A local account created for recovery should not automatically receive unrestricted administrative access.
Similarly, a service identity created for automation should not receive the full rights of a platform administrator when its job requires only a small number of API operations.
Flow Network Security Is a Workload Protection Control
Nutanix Flow Network Security supports application-centric microsegmentation and policy-driven workload isolation in supported AHV environments. See the Flow Network Security overview.
It can help restrict application communication and reduce lateral-movement paths.
Its actual availability depends on the deployed software, supported configuration, and licensing entitlement.
Flow Network Security should not be treated as a universal replacement for physical management-network isolation.
Nor should its presence imply that every VM or control-plane component is automatically protected by an appropriate policy.
The design needs an application map, approved security policy, change ownership, and validation of the resulting traffic behavior.
The team must also define who can change policy during an incident.
If the same compromised privileged identity controls both workload administration and segmentation, the security boundary may be weaker than its network diagram suggests.
Protect Cluster Administration and Supporting Infrastructure
Prism Central, Prism Element, AHV hosts, and CVMs are not interchangeable administrative endpoints.
The access paths and privileges needed for routine VM operations differ from those needed for platform maintenance, storage administration, and emergency recovery.
A sound design restricts each interface to approved management networks and identities, maintains the supported hardening baseline, and monitors critical administrative activity.
BMC and physical network infrastructure must remain in scope.
A virtual network policy cannot compensate for an exposed hardware management controller that allows a separate attacker to alter server state.
Secure Snapshots Are Not the Same as Trusted Recovery
Nutanix provides data protection and disaster recovery capabilities, including secure or immutable snapshot options and replication mechanisms. See Nutanix’s security capabilities.
Those capabilities can protect selected recovery points against certain forms of alteration or deletion when appropriately configured.
They do not establish that a snapshot contains clean application data or trustworthy administrative configuration.
An attacker may have modified a workload before the protected recovery point was created.
The recovery design must therefore address copy protection, management credentials, retention, restore dependencies, and integrity validation separately.
Prism Central configuration and recovery state also require protection. Depending on the deployed services and release, not every configuration object or self-service artifact necessarily shares the same backup and restoration behavior. See the Prism Central resilience guide.
The exact backup coverage and supported recovery procedure should be validated against the installed Prism Central version.
Compare Responsibilities, Not Product Names
The three platforms do not provide identical control models.
The comparison below focuses on decisions an architect must make before designing containment and recovery.
| Architecture dimension | VMware Cloud Foundation | Azure Local | Nutanix |
|---|---|---|---|
| Central management | VCF Operations, VCF Management Services, SDDC Manager and related components | Azure management and Arc-integrated services for supported connected operations | Prism Central |
| Local virtualization administration | vCenter and ESX management | Azure Local cluster, physical-machine and supported local management interfaces | Prism Element, AHV and related cluster services |
| Identity dependencies | VCF Identity Broker, external IdP and local platform identities | Microsoft Entra for Azure authority, with separate local infrastructure and workload identity dependencies | Prism Central authorization, directory integration, local and service identities |
| Workload segmentation | NSX and applicable vDefend policies | Azure Local NSGs for supported Azure-managed VMs, plus other supported local or external network controls | Flow Network Security for supported deployments and external network controls |
| Management isolation | Restricted VCF management and ESX networks, supporting firewalls and administrator paths | Separate cloud, local management, hardware and workload boundaries | Restricted Prism, cluster, CVM, host and hardware administration |
| Evidence | VCF component audit records, platform logs, independent collectors | Azure activity, local infrastructure and workload logs, supported independent storage | Prism, cluster and workload records, independent collectors |
| Recovery considerations | Restore trust in management components, identity, NSX policy and supporting infrastructure | Restore cloud/local management relationships and supported Arc dependencies | Restore Prism and cluster authority, policies, required keys and workload services |
| Primary design caveat | Integrated management can concentrate administrative reach | Cloud management and local infrastructure have different failure and recovery conditions | Central Prism administration and cluster-level authority require separate evaluation |
This table is not a feature-parity scorecard.
It identifies where ownership and controls belong.
The relevant question is not which product has the longest security feature list. It is whether the deployed combination satisfies the organization’s mandatory trust and recovery requirements.
Apply Hard Gates Before Selecting the Architecture
Security design can become misleading when a platform receives a high weighted score despite failing one critical requirement.
Use mandatory gates first.
A proposed management or recovery architecture should be rejected or redesigned if it cannot demonstrate the conditions below.
| Hard gate | Required evidence | Consequence if absent |
|---|---|---|
| Restricted management reachability | Network path inventory and approved enforcement test | Workload or user compromise may reach infrastructure administration |
| Bounded privileged authority | Effective permissions, identity ownership, elevation procedure | One credential may modify unrelated resources or protections |
| Independent containment | Authorized method to restrict compromised access | Incident response depends on the attacker-controlled plane |
| Protected evidence | Demonstrated audit export, integrity and retention controls | Compromise may prevent reliable reconstruction |
| Recovery access | Trusted identities, connectivity, tools and required keys | Backups exist but cannot be safely restored |
| Trusted recovery source | Verified configuration, software and recovery-point selection | Recovery may reproduce malicious state |
| Business-service acceptance | Defined application and operational checks | Restored infrastructure may remain unusable or unsafe |
These gates should be evaluated against realistic scenarios, not design diagrams alone.
A platform may be suitable for a particular service while another deployment of the same product fails because it has excessive standing privileges, flat networking, missing evidence, or no trusted recovery path.
The Persistence Paths an Architecture Review Must Expose
A security architecture should reveal not only where an attacker might enter, but which access can remain after the original entry point is removed.
Four patterns deserve particular attention.
Additional Administrative Identities
A compromised administrator may create another local or federated identity, grant a role, or register a new application credential.
Restricting the original identity does not automatically remove the new authority.
For VCF, review relevant vCenter, SSO, host, and management-service identities. For Azure Local, examine Azure role assignments alongside local and infrastructure identities. For Nutanix, review Prism Central authorization and applicable cluster-level accounts.
The response must identify the actual authority store involved. There is no single cross-platform account list that contains every administrative identity.
Management Appliance Persistence
An attacker may alter startup configuration, install an unauthorized service, or exploit available management interfaces.
Traditional guest endpoint security may not inspect the underlying appliance or hypervisor.
Mandiant’s BRICKSTORM investigations demonstrate this monitoring gap in VMware management environments.
For all three platform families, the preventive design should inventory management appliances, restrict unnecessary access, preserve supported configuration baselines, and collect independent network and administrative evidence.
Do not assume an endpoint agent can be installed on every platform component. Use the supported controls and external visibility mechanisms appropriate to each appliance or host.
Automation and Configuration Reconciliation
Automation can recreate previously approved resources after a failure or redeployment.
That is useful during ordinary operations.
During compromise, an altered desired-state repository, deployment identity, workflow, or configuration source may recreate unauthorized access.
The security review should include automation that manages users, network policies, virtual machines, backup jobs, and platform configuration.
Recovery should not automatically resume every pending workflow until the intended state and accepted downstream changes have been reconciled.
Recovery-System Authority
Backup and recovery services frequently require broad access to production infrastructure.
An identity that can delete recovery copies, modify retention, access encryption keys, or alter restore destinations may be more consequential than one that can administer an ordinary workload.
Separate routine backup operation from high-impact recovery-policy administration wherever supported.
Require independent evidence for changes to copy retention, backup destinations, encryption keys, and recovery access.
The objective is to preserve recovery choices even when some production credentials have been compromised.
Design a Control Boundary That Can Survive Compromised Management
The key design principle is independent enforcement.
A network policy managed through the same compromised interface it protects may still be useful, but its administrative independence is limited.
This matters differently across the three platforms.
In VCF, a critical management perimeter can include independently administered physical firewall rules as well as supported NSX controls.
In connected Azure Local, Azure-managed networking and identity controls should be assessed alongside local infrastructure access, physical networking, and any separate on-premises SDN mechanisms.
In Nutanix, Flow Network Security may protect application traffic, while external network controls continue protecting Prism, cluster administration, and supporting infrastructure.
The design should not duplicate every control for its own sake.
It should identify which control must remain effective when another control or its administrator becomes untrusted.
A useful dependency model is:

Independence also creates cost and complexity.
A separate firewall, recovery vault, identity path, or administrative environment requires lifecycle management and operational ownership.
The justification should be tied to the impact of losing the original management plane, not to an abstract requirement for additional security products.
Make Recovery Copies Useful After a Compromise
Replication and backup design should distinguish availability from trust.
Replication may faithfully reproduce corrupted files, malicious configuration, or unwanted application changes.
A protected snapshot can preserve a recovery point, but its contents may already reflect an unauthorized state.
An immutable retention mechanism may prevent certain deletion or alteration operations while leaving other risks unresolved, such as compromised restore credentials or malicious configuration inside the protected copy.
A trusted recovery design therefore needs several related decisions.
Protect Copies Independently
Identify which identities can change retention, delete copies, alter backup destinations, or administer the recovery platform.
Separate those privileges from ordinary workload administration where supported.
Verify the protection mechanism actually applied to the chosen repository or snapshot policy.
Do not assume that every backup feature marketed as secure or immutable provides identical retention guarantees.
Preserve the Configuration Needed to Restore
A VM backup may not contain everything needed to restore a functioning business service.
The service may depend on virtual networking, DNS, certificates, identity, encryption keys, management-plane services, application configuration, and external integrations.
Each platform has its own procedures and dependencies.
The recovery architecture should therefore define the complete service restoration boundary rather than relying solely on VM-level copy availability.
Establish a Trusted Recovery Candidate
Before releasing a restored system, validate the selected recovery point and the administrative environment used to restore it.
Where compromise is suspected, compare configuration against approved baselines and investigate whether unauthorized access was introduced before the recovery point.
Restore required authority deliberately rather than accepting every existing credential, service account, or automation job as trustworthy.
Preserve the Ability to Withhold Service
A critical recovery control is the ability to keep a restored system isolated until the investigation and business checks are complete.
That may require a separate recovery network, restricted DNS or routing, controlled identity services, and clear approval to reconnect external integrations.
The precise implementation should use the supported recovery mechanisms for the platform and workload.
This article defines the requirement; the next installment will examine the operational sequence after management-plane compromise.
A Practical Cross-Platform Design Example
Consider a hypothetical customer-order platform distributed across three infrastructure estates.
The primary application runs on VCF. A warehouse processing service runs on Azure Local. An analytics service runs on Nutanix.
The systems exchange order status, inventory, and reporting data through approved APIs.
The infrastructure teams use a shared identity provider and centralized security monitoring. Different teams own the underlying platforms.
An attacker compromises an infrastructure automation credential and attempts to gain persistent administrative access.
The service architecture should answer several questions before the incident occurs.
Can the attacker reach every platform? Determine whether the credential is scoped to one platform, one environment, or several independent management systems.
Can the attacker change protective policy? Identify whether the identity can modify NSX rules, Azure-managed network security groups, Flow policies, or external firewalls.
Can the attacker access recovery? Determine whether the same credential can modify backup jobs, delete retained copies, retrieve encryption keys, or authorize a restore.
Can investigators retain evidence? Identify whether administrative logs are forwarded to protected storage outside the affected platform.
Can service owners continue operating? Establish whether the warehouse can continue in a reduced mode, whether the ordering service can stop unsafe writes, and how the analytics system behaves when an integration is restricted.
These are cross-platform architecture questions.
They do not require pretending that all three vendors use the same administrative model.
The correct design may use different native controls on each platform while standardizing the surrounding decisions: identity scope, independent enforcement, evidence, containment authority, and business acceptance.
A Phased Implementation Path
An organization does not need to redesign its entire infrastructure estate at once.
Begin with one critical business service and the management paths that can affect it.
| Phase | Engineering work | Required output | Exit condition |
|---|---|---|---|
| Discover | Map workloads, management systems, privileged identities, automation and recovery dependencies | Current-state trust and service map | Critical unknowns have assigned owners |
| Define boundaries | Select independent administrative, network, evidence and recovery controls | Approved architecture and decision records | Mandatory security gates are met in design |
| Implement | Apply supported platform hardening, segmentation and identity restrictions | Version-specific configurations and change evidence | Controls enforce the intended scope |
| Exercise | Simulate a compromised identity or management dependency in an authorized environment | Denial, recovery and business-validation evidence | Containment and recovery behavior meet accepted criteria |
| Operate | Monitor drift, privileged changes, recovery readiness and lifecycle | Service-level security and resilience measures | Exceptions are owned and retested |
The phase gates are more important than the calendar.
If an organization cannot demonstrate independent administrative access during a management failure, the design should not be declared recovery-ready simply because its documentation is complete.
Keep the Design Contract Machine-Readable
The following YAML is a conceptual architecture contract. It is not an executable configuration for VCF, Azure Local, or Nutanix.
Its purpose is to make the intended trust boundaries and required evidence explicit.
persistent_compromise_design:
service: critical-business-service
architecture_status: proposed
owners:
business: service-owner
security: incident-response-lead
platform: infrastructure-operations
identity: identity-services
recovery: recovery-engineering
protected_platforms:
- vmware-cloud-foundation
- azure-local
- nutanix
management:
direct_internet_access: prohibited
approved_admin_sources: required
vendor_access: time_bound_and_logged
hardware_management_isolated: true
identity:
shared_privileged_accounts: prohibited
scoped_automation_identities: required
emergency_access_independently_protected: true
authority_change_audit: required
enforcement:
workload_boundaries: approved_per_platform
critical_management_boundary:
independent_administration: required
unauthorized_policy_change:
detection: required
restriction_path: documented
evidence:
administrative_events: required
independent_retention: required
configuration_baselines: required
protected_time_source: required
recovery:
protected_copies: required
trusted_admin_path: required
required_keys_available: required
isolated_validation: required
queued_automation_reconciled: required
business_acceptance: required
release_gates:
unauthorized_access_denied: required
recovered_configuration_validated: required
business_transactions_verified: required
The values are design intent.
Each requirement must be translated into the appropriate supported product configuration, assigned an owner, and given an observable validation method.
A Boolean field stating that emergency access is protected is not evidence that emergency access has been tested.
For a production implementation, associate each control with its actual policy, procedure, test record, and exception history.
Test Containment Without Using a Live Incident as the First Exercise
A cyber-resilience architecture should be tested under safe, authorized conditions.
The exercise can be performed in a representative nonproduction environment or through carefully approved production tests that do not endanger critical services.
The objective is to validate specific trust boundaries, not to simulate uncontrolled destructive behavior.
Scenario: A Privileged Automation Identity Is Compromised
Select a test identity with a defined, limited administrative scope.
Attempt approved operations against resources inside and outside that scope.
Record whether the unauthorized operations are denied by the target platform or independent enforcement layer.
Then simulate restriction of the identity through the intended incident procedure.
The test should establish that the restriction is effective for new operations and account for applicable existing sessions or issued credentials.
Do not assume every platform uses identical token, session, or revocation behavior.
Scenario: Management Connectivity Is Lost
Exercise the loss of an administrative dependency.
For VCF, that might involve a selected vCenter management dependency in an isolated environment.
For connected Azure Local, it might involve restricted connectivity to an Azure management dependency while preserving the local platform’s supported operating conditions.
For Nutanix, it might involve loss of Prism Central connectivity while validating the documented cluster-level operating and management boundaries.
These are proposed test cases, not claims of measured product behavior.
The expected result must be established from the exact versions and supported architecture before execution.
Scenario: A Recovery Point Contains Untrusted State
Prepare a nonproduction workload with known baseline configuration and an intentionally introduced test deviation.
Use the cyber-resilience executive metrics to connect the exercise evidence to recovery and containment outcomes.
Restore the candidate recovery point into an isolated validation environment.
Determine whether the recovery procedure can identify the deviation before the restored service is approved to reconnect.
The test should examine the relevant identity, configuration, application, and data-integrity checks.
A successful boot is not sufficient.
Record the Evidence That Matters
| Test objective | Observable result | Failure condition |
|---|---|---|
| Management isolation | Unauthorized path is denied | Workload or user network reaches protected management service |
| Privileged access restriction | Unapproved operation is rejected | Identity can exceed its approved scope |
| Independent containment | Authorized responder can restrict affected access | Restriction depends exclusively on compromised authority |
| Evidence preservation | Required audit events remain available independently | Relevant records are inaccessible or alterable through affected authority |
| Recovery administration | Trusted recovery path is usable | Recovery requires unavailable or untrusted credentials |
| Recovery integrity | Approved configuration and application state are verified | Malicious or unexplained state is accepted without review |
| Business acceptance | Representative service transactions pass | Infrastructure is restored but the service is unusable |
Targets should reflect service criticality and actual operating requirements.
Do not invent a universal containment-time objective or assume all environments can meet the same recovery duration.
Record observed times and their test conditions separately from planning targets.
Day-2 Operations Must Protect the Boundaries After Deployment
The architecture will change.
Administrators receive new responsibilities. Automation expands. Workloads move. Vendors introduce software updates. Network policies accumulate exceptions.
A design that begins with strong isolation can become fragile if those changes are not governed.
Review Effective Authority, Not Just Account Counts
An identity inventory is a starting point.
The higher-value review establishes what each identity can actually change.
Prioritize accounts and service identities that can modify platform administrators, network policy, recovery configuration, encryption keys, or software distribution.
A small number of highly privileged identities may represent more concentrated risk than hundreds of narrowly scoped workload accounts.
Monitor Management-Plane Drift
Each platform provides supported hardening and configuration-management capabilities.
Broadcom documents security configuration guidance and VCF Operations visibility for VCF 9.1.
Microsoft documents Azure Local security baselines and supported drift-control behavior.
Nutanix documents Prism security roles, cluster hardening, and lifecycle controls.
Those capabilities require an operating process.
An unexpected configuration change should be attributable to an approved change, authorized administrator, or documented exception.
During a suspected compromise, drift should not be automatically overwritten before the relevant evidence has been preserved.
Protect the Lifecycle Process
Infrastructure updates affect management components, authentication, network policy, certificates, and recovery dependencies.
Use the supported lifecycle method for each platform.
Do not bypass platform management requirements with ad hoc changes simply because the underlying operating system accepts the command.
For security-driven updates, define prerequisites, maintenance dependencies, stop conditions, rollback or recovery limitations, and post-change validation.
The fact that a patch can be installed with little workload disruption does not eliminate the need for operational evidence.
Revalidate After Material Changes
Revisit the trust design when a new management integration is introduced, a workload domain or cluster is added, a privileged identity changes, or a recovery platform is replaced.
Include changes to cloud connectivity, backup retention, hardware management, and supplier access.
The objective is not to repeat a full security assessment for every minor change.
It is to recognize changes that affect the authority, independence, or recoverability of critical infrastructure.
Common Architectural Failure Modes
Several designs create a false sense of resilience.
A management VLAN is mistaken for isolation. Routing, broad firewall rules, or a shared jump host still permit uncontrolled access.
A centralized identity provider is mistaken for independent administration. All emergency access depends on the same provider and privileged workstation.
Microsegmentation is assumed to protect its own administrator. Workload policy may be changed by an attacker who controls the relevant management plane.
A healthy Azure Local VM is mistaken for a healthy Azure management path. Existing workloads may continue while Arc-dependent operations are unavailable or compromised.
Prism Central visibility is mistaken for complete cluster control evidence. Cluster-level identities, CVMs, hosts, and supporting infrastructure still require investigation.
A protected snapshot is mistaken for a clean recovery point. Unauthorized configuration may already exist inside the retained state.
Backup infrastructure shares production administrator credentials. The same compromised authority can affect both live workloads and recovery options.
Monitoring is centralized but not independent. An attacker with management access may interfere with the records needed to reconstruct the incident.
Automation restarts before reconciliation. Previously queued or modified workflows may resume using newly restored authority.
A platform feature comparison replaces engineering validation. The organization chooses products based on advertised capabilities without demonstrating the actual trust and recovery boundaries.
These failure modes are not reasons to avoid VCF, Azure Local, or Nutanix.
They are reasons to design security controls around administrative authority, platform dependencies, and verifiable operating outcomes.
Conclusion
Persistent compromise changes how enterprise infrastructure architecture should be evaluated.
The objective is not simply to keep virtual machines available during an infrastructure failure. It is to preserve trusted administration, enforce meaningful boundaries, retain independent evidence, and recover business services when one or more management components can no longer be trusted.
VMware Cloud Foundation, Azure Local, and Nutanix approach infrastructure management differently.
VCF integrates multiple management, control, and workload components. Connected Azure Local combines cloud-based governance with locally operated infrastructure. Nutanix separates centralized Prism management from cluster-level platform services.
Those differences matter because containment and recovery must follow the real control relationships.
No single platform security feature replaces the need to understand who can alter authority, what an attacker can reach, which protections can be changed from a compromised account, and how the environment can be restored independently.
Start with one critical service and apply the hard gates in this article.
Map its management dependencies. Restrict privileged reach. Protect evidence and recovery authority. Test containment through a trusted path. Restore only after the required configuration and business outcomes have been validated.
The architecture is resilient against persistent compromise only when the enterprise can remove untrusted authority without losing the independent control required to investigate and recover.
These design conditions support the companion guide on cyber recovery after management-plane compromise.
External References
- Google Threat Intelligence Group and Mandiant: Another BRICKSTORM: Stealthy Backdoor Enabling Espionage into Tech and Legal Sectors
- NIST: Developing Cyber-Resilient Systems: A Systems Security Engineering Approach, SP 800-160 Volume 2 Revision 1
- VMware Cloud Foundation Blog: Good Fences, Good Neighbors: Isolating Infrastructure Management
- VMware Cloud Foundation Blog: Securing your VMware Cloud Foundation 9.1 Environment
- VMware Cloud Foundation Blog: Strengthen Zero Trust Security and Resilience with VCF 9.1
- Microsoft Learn: Architecture best practices for Azure Local
- Microsoft Learn: What is Azure Arc resource bridge?
- Microsoft Learn: Azure Arc resource bridge maintenance
- Microsoft Learn: Manage security defaults for Azure Local
- Microsoft Learn: Create network security groups on Azure Local
- Nutanix: Security Guide (select the installed release in the Support Portal)
- Nutanix: Enterprise Network Security for Multicloud Apps and Data, Flow Network Security
- Nutanix: Secure Your Workloads and Data
- Nutanix: Fine-Grained RBAC with Prism Central
- Nutanix: Prism Central Resilience