Storm-0940 and ConnectWise: How Stolen Application Trust Reaches Customers

TL;DR

Microsoft’s October 21, 2026 investigation attributes a campaign involving compromised ConnectWise Cloud Backup application identities to Storm-0940, a China state-aligned threat actor. The attacker used the ARTEX agentic penetration-testing framework to accelerate authentication attempts against more than 2,250 Microsoft 365 customer tenants. That figure describes attempted tenant reach, not 2,250 confirmed breaches. Microsoft observed follow-on Microsoft Graph and directory activity in some customer environments, including credential changes and attempts to expand privileged access.

The underlying exposure was a trusted application relationship. A multitenant supplier application can authenticate through an existing service principal in a customer tenant, exercising permissions previously granted to that application. Revoking the supplier’s compromised credentials restricts new authentication, but does not automatically remove unauthorized accounts, additional credentials, grants, or other access established inside customer environments.

The practical requirement is to investigate supplier trust and customer authority separately. Identify the affected application identities, determine which tenants were actually accessed, investigate downstream changes, and restore service only after security restrictions and business operations have been independently validated.

In This Article

Introduction

Consider a hypothetical managed service provider responsible for Microsoft 365 backup and recovery across dozens of customer environments.

Each customer has authorized a cloud backup application. The integration runs without an employee signing in every time a scheduled backup occurs. The supplier maintains the application, and customers depend on it to protect Exchange Online, SharePoint, OneDrive, and other supported Microsoft 365 data.

Then the supplier announces that application credentials have been compromised.

The provider disables the affected connections. Security teams begin reviewing authentication records. Customers want backups running again. The supplier is investigating its own environment, but the customer still needs to determine whether the compromised application performed unauthorized operations inside the tenant.

This is the architectural problem behind the October 2026 ConnectWise incident.

Microsoft reports that Storm-0940 used ARTEX to accelerate activity through applications already trusted by customer organizations. The attack did not depend on discovering a new vulnerability in Microsoft Entra or Microsoft Graph. It abused authentication material and permissions associated with an established integration.

For enterprise architects, managed service providers, identity teams, and security leaders, the incident creates a difficult balance.

The compromised connection must be restricted. The customer environment must be investigated. Backup availability must eventually be restored. However, reconnecting before understanding what access survived could reintroduce risk, while leaving backups unavailable creates its own operational exposure.

This article extends the series’ enterprise risk model into the documented supplier-to-customer attack path, its limits, and the decisions needed to contain and recover.

It focuses on ConnectWise Cloud Backup, formerly SkyKick, and the associated application identities. The separate ScreenConnect and SCKit activity belongs to another article in this series. Shared supplier branding does not establish a common technical intrusion.

What Microsoft and ConnectWise Actually Confirmed

Microsoft published its Storm-0940 investigation on October 21, 2026.

It describes Storm-0940 as a China state-aligned threat actor active since 2021, historically associated with password spraying, exploitation of internet-facing applications, and abuse of compromised cloud identities.

Microsoft observed the actor using ARTEX as early as September 2026.

ARTEX is an open-source, multi-agent penetration-testing framework capable of coordinating reconnaissance, tool execution, and later intrusion activities. Microsoft identified multiple attacks in which privileged cloud access was obtained less than three hours after initial targeting.

That is an observed outcome in particular attacks, not a guaranteed performance characteristic of the framework.

The Relevant Timeline

DateDocumented developmentOperational significance
September 2026Microsoft observed Storm-0940 using ARTEX in attacks across multiple industries.AI-assisted intrusion orchestration had entered observed operations.
October 10–14Microsoft observed a separate password-spraying campaign targeting 327 accounts at eight organizations.More than 151,000 failed sign-ins were recorded, without successful authentication in that wave.
October 14ConnectWise reports discovering unauthorized access involving Cloud Backup application identities.Supplier investigation and containment planning began.
October 16ConnectWise published its initial Cloud Backup advisory.Partners received notice of the affected trust relationship.
October 17ConnectWise disabled seven application identities and revoked associated credentials.Vendor-side containment was applied, but customer-side changes still required review.
October 21Microsoft published the Storm-0940 investigation.Public attribution and detailed downstream activity became available.
October 28ConnectWise began restoring Cloud Backup through reauthorization using a new application.Customer administrators needed to restore authorization deliberately.
November 4ConnectWise published its latest Cloud Backup restoration update in the reviewed advisory.Backup and restore capabilities were available for reauthorized tenants; some secondary functions were still being restored.
November 6Okta published related research into large-scale identity-flow reconnaissance.Shared infrastructure and some overlapping activity were documented, without establishing a single actor behind every observed event.

These events must not be treated as one continuous set of confirmed successful intrusions.

The password-spraying activity and the subsequent application-trust abuse are separate observations, even though Microsoft connected relevant infrastructure to its Storm-0940 investigation.

The 2,250-Tenant Number Is Not a Breach Count

Microsoft reported that Storm-0940 used ARTEX to attempt sign-ins against more than 2,250 customer tenants following the compromise of ConnectWise application credentials.

A subset of the activity involved rapid authentication with several compromised application credentials.

In some environments, the actor performed additional Microsoft Graph operations, including reconnaissance, application credential modifications, and changes involving identities, groups, or privileged roles.

Those findings are consequential, but they describe different levels of confirmed activity.

FindingWhat it establishesWhat it does not establish
A tenant contained an affected applicationPotential exposure through an existing trust relationshipThat the attacker used it
An attempted authentication occurredAn identity was tested against the tenantSuccessful access
Authentication succeededThe identity was accepted for a resourceData theft or privilege changes
A Graph operation was observedAn API request was attempted or performed, depending on the resultEvery requested operation succeeded
A credential or role was changed without authorizationConfirmed follow-on activity and possible persistenceCompromise of every related system
Data access or transfer was confirmedA documented resource-level impactIdentical impact across other tenants

Microsoft explicitly cautioned that an affected application or successful authentication does not necessarily establish customer-data access or exfiltration.

That distinction should remain visible in both incident reports and executive communications.

How a Trusted Supplier Application Reaches Customer Resources

To understand the incident, it helps to separate four components that are often called simply an “application.”

The Supplier’s Application Registration

The software supplier maintains a multitenant application registration in its home Microsoft Entra tenant.

That registration defines the application identity and supported authentication configuration.

The supplier manages credentials used by its application, such as client secrets, certificates, or supported federated identity credentials.

In the ConnectWise incident, the compromised authentication material was associated with supplier-managed applications.

The Customer’s Service Principal

When the multitenant application is authorized in a customer tenant, Microsoft Entra creates a corresponding enterprise application represented by a service principal.

This is the customer’s local instance of the trusted application relationship.

The application ID identifies the application across tenants. The service-principal object ID identifies the specific object inside one tenant.

Those IDs are not interchangeable.

An investigation that searches only by application display name can miss older SkyKick names, renamed integrations, or multiple objects with similar descriptions.

The customer grants the application permissions needed to operate.

Depending on the integration, those permissions may allow application-only access to APIs and data without requiring a human user to approve every operation.

That is normal for background services such as backup, monitoring, and automation.

The important question is whether the effective permissions exceed the application’s current business purpose, and what becomes possible if the supplier’s authentication credentials are stolen.

Microsoft Graph and Downstream Resources

With an accepted application identity and sufficient authorization, the caller can invoke permitted Microsoft Graph APIs.

The same trust relationship may also extend to Azure resources or other services when the application has been granted relevant access.

Microsoft’s investigation documented follow-on Graph activity and, in its Defender for Key Vault coverage, activity involving secret reads across multiple Azure Key Vaults, key enumeration, and a key-unwrapping operation.

These observations do not establish that every affected ConnectWise tenant had Azure Key Vault exposure. They demonstrate why a trusted application requires a resource-level investigation rather than a sign-in review alone.

The Trust Dependency Map

The diagram below separates the supplier’s credential boundary from the customer tenant’s authorization boundary.

The architectural lesson is that a supplier-side credential and a customer-side authorization grant are separate controls.

ConnectWise can revoke credentials it manages and disable the affected supplier application identities.

That does not automatically prove that every corresponding service principal was disabled in every customer tenant, every previously issued token was rejected immediately, or every unauthorized downstream change was reversed.

ConnectWise’s technical guide explicitly states that its vendor-side actions did not change local customer enterprise-application configurations.

A successful supplier containment action is therefore necessary evidence, but it is not sufficient evidence of customer-tenant eradication.

Which ConnectWise Application Identities Must Be Reviewed?

ConnectWise’s October 18 technical investigation guide identifies seven application identities disabled on October 17.

These identifiers are useful because historical installations may still display SkyKick names.

ApplicationApplication client ID
ConnectWise Cloud Backup506e08f9-6e9a-4da4-942b-a099a03644ae
ConnectWise Cloud Management2399d300-6fca-4e6a-a686-399fefe75b83
Cloud Management – MS Teams24205314-2b89-4e07-a2a4-579e898a5640
Cloud Management – Exchange Onlineb78f6a77-d4e5-46fb-87e5-d1648bd02519
Cloud Management – MS Graph0de14c77-b9b0-4137-9360-34878a929827
Cloud Management – Partner Center412dcf96-176d-4b3a-a1b0-4981ad66971e
Cloud Management – SharePoint Onlineb2b55f5b-7758-4719-8d8f-88ae172e6bdb

These are the vendor-published identifiers for the affected-application review. They are not evidence that every listed application was used maliciously in every tenant.

Microsoft’s threat-hunting examples focus on a smaller subset of relevant application IDs. For a comprehensive tenant inventory, use the full supplier-published list, then match findings to Microsoft telemetry and the actual application configuration.

For each match, record the tenant ID, application ID, local service-principal object ID, current status, granted permissions, and relevant investigation results.

Do not confuse the supplier-owned application registration with a locally registered application. Customer administrators generally cannot inspect or rotate credentials held in the supplier’s home tenant.

Their responsibilities concern local enterprise application state, permissions, audit evidence, and any unauthorized local identities or changes.

Why Application-Only Access Changes the Security Model

Many identity programs are designed around interactive users.

They emphasize multifactor authentication, password policies, device compliance, conditional access, privileged access management, and user sign-in monitoring.

Those controls remain important.

But the ConnectWise case illustrates a different access pattern.

In application-only authentication, a service principal authenticates using application credentials. A person does not necessarily sign in during the transaction.

The resulting access depends on the granted permissions and the resource’s enforcement of those permissions.

That means a security team can have strong interactive-user multifactor authentication and still face an application-credential compromise.

The Problem With Standing Permissions

A backup product may legitimately require broad access to protected data.

However, broad permissions create a significant dependency on the supplier’s credential security and application lifecycle.

A security review should determine whether permissions are necessary for the current service, whether less privileged alternatives exist, and whether some administrative operations can be separated from routine backup activity.

For example, an application that retrieves backup data should not receive unrelated directory administration authority merely because it is convenient during onboarding.

Where privileged access is unavoidable, the organization should understand how it can be restricted during an incident and which business functions will fail when that happens.

Credential Rotation Does Not Automatically Remove Persistence

Microsoft documented that Storm-0940 has previously established persistence by adding credentials to existing OAuth applications.

In the ConnectWise investigation, Microsoft also observed application credential modifications and changes involving identities and roles in downstream environments.

That is the reason revocation must be followed by a change-history investigation.

Imagine a compromised application that has permission to manage another application.

The attacker uses the first application’s valid credentials, creates a second credential on another identity, and then loses access when the supplier rotates the original secret.

The second credential may still be valid.

This is an illustrative scenario consistent with the reported persistence pattern, not a claim that every customer tenant experienced that exact sequence.

The response must examine authority created during the compromise window, not just the original credentials that were reported stolen.

What Microsoft Observed Through Microsoft Graph and Azure Key Vault

Microsoft Graph provides a consistent API for interacting with Microsoft 365 and Microsoft Entra resources.

For legitimate applications, this reduces integration complexity. For attackers with a compromised authorized identity, it can provide an efficient way to enumerate or modify resources within the application’s effective permissions.

Microsoft reported suspicious downstream activity including application credential changes, reconnaissance, and attempts to alter privileges.

It also described Defender detections for unusual mailbox access, SharePoint enumeration, high-privilege application-role assignments, and related behaviors.

The presence of a detection category is not proof that every behavior occurred in every affected tenant. Each finding must be supported by the corresponding records.

Key Vault Requires Its Own Resource Investigation

Microsoft’s report separately documented activity involving Azure Key Vault.

The reported behavior included high-volume secret access across multiple vaults, key enumeration, and a key-unwrapping operation.

A compromised application able to access secrets may create a much wider downstream investigation requirement than its initial Microsoft 365 purpose suggests.

Secrets retrieved from a vault can potentially authenticate to other systems if the attacker obtains usable values and those secrets remain valid.

The appropriate response is not automatically to rotate every enterprise secret.

Instead, identify the vaults and resources the affected identity could access, determine which operations occurred, establish whether secret values were retrieved, and prioritize credential replacement according to confirmed or credible exposure and business dependencies.

Preserve relevant Key Vault logs and coordinate changes with the owners of affected applications.

A successful credential rotation may interrupt an application unless the replacement has been distributed through a trusted deployment process.

That dependency should be part of the incident plan before revocation begins.

What Okta’s Independent Research Adds

Okta Threat Intelligence published its investigation on November 6, 2026.

The report describes large-scale reconnaissance, fraudulent account registration, and attempted intrusion activity across hundreds of Okta and Auth0 tenants.

Okta observed attackers probing public identity endpoints, registration behavior, authorization flows, credential-stuffing opportunities, and authentication configurations.

The researchers also identified overlap between some of the infrastructure reported by ConnectWise and Microsoft and infrastructure used during activity against Okta customers.

The shared addresses included 144.172.64.60, 144.172.64.99, and 144.172.64.128.

That overlap is meaningful investigative evidence.

It does not establish that every event observed by Okta was performed by Storm-0940.

Okta assessed that multiple actors had almost certainly tasked one or more agentic frameworks with identity-focused operations, likely involving ARTEX or related variants. Okta explicitly stated that it could not independently confirm that every observed operation used ARTEX.

Its assessment was that Storm-0940 was one of several operators using this class of tooling.

This illustrates a useful attribution boundary.

Microsoft’s Storm-0940 label applies to a campaign Microsoft tracked. Google and Wiz reported overlapping ARTEX-associated activity under different tracking conventions. Okta observed infrastructure and behavioral overlap without collapsing every identity attack into one actor.

The same framework, infrastructure provider, or month of activity does not independently prove common command.

No Identity-Platform Vulnerability Was Established

Okta also reported that its observations did not establish exploitation of vulnerabilities or misconfigurations specific to Okta, Auth0, Microsoft Entra, or Amazon Cognito as identity platforms.

The attacks largely involved familiar identity weaknesses, including registration flows, weak authentication configurations, and credential-stuffing opportunities.

The agentic contribution was the ability to test and adapt those methods at scale.

For defenders, the lesson is to examine existing identity exposure and control effectiveness instead of assuming that installing an AI security product resolves the problem.

Investigating Each Customer Tenant

The response must distinguish supplier containment from customer-tenant investigation.

ConnectWise disabled its seven affected application identities and revoked associated credentials at 07:00 UTC on October 17, 2026.

That action did not reverse changes already made inside customer tenants.

ConnectWise also stated that its Cloud Backup environment did not provide the downstream security visibility needed to establish whether a specific customer experienced unauthorized activity.

Each affected customer environment therefore needs its own investigation, even when the supplier reports successful containment.

Identify the Correct Tenant and Application

Begin with the customer tenant ID and the seven supplier-published application IDs.

Review both currently authorized applications and historical SkyKick or ConnectWise integrations.

For every identified enterprise application, record the application ID, tenant-specific service-principal object ID, enabled status, owners, and granted permissions.

Preserve the current configuration before destructive cleanup. Do not delay urgent containment while waiting for evidence exports.

The investigation should use a trusted administrative path and appropriate roles. Reading authentication records, managing applications, investigating Exchange activity, and modifying privileged roles require different permissions.

Do not assign Global Administrator simply because a broad investigation worksheet contains several different tasks.

Search the Relevant Authentication History

ConnectWise’s technical guide recommends reviewing activity from October 13, 2026, with broader scoping to October 1 where available.

Microsoft’s published hunting example starts on October 13, 2026. Preserve earlier records where available and extend the investigation when the evidence supports it.

For a comprehensive investigation, use the earlier applicable period and widen it when evidence supports doing so.

Record the timestamps in UTC or preserve the source timezone and offset.

The important complication is retention.

As of November 18, 2026, many October events are already older than the default Microsoft Entra sign-in and audit-log retention periods.

Microsoft documents seven days for Entra ID Free and 30 days for Entra ID P1 or P2 for these record types. Longer retention may exist through previously configured export, Azure storage, Microsoft Sentinel, a security information and event management (SIEM) system, or separately retained Microsoft 365 audit records.

Upgrading a license now does not recover records that have already expired.

The investigation must therefore record its actual visibility period rather than claiming to have searched a period for which no telemetry remains.

Review Service-Principal Sign-ins

Service-principal sign-ins are separate from interactive and noninteractive user sign-ins.

Microsoft documents that service-principal records can represent authentication using certificates or application secrets, including OAuth client-credentials flows.

These records can be grouped when the service principal, status, IP address, and resource are the same.

Expand grouped events before interpreting their timing or frequency.

A useful review compares successful and failed authentication by application ID, source address, target resource, and available correlation identifiers.

The published suspicious addresses are relevant investigation leads, but they are not a complete list of possible attacker infrastructure.

An event from an unfamiliar IP address is not automatically malicious. Conversely, an event from a familiar provider address is not automatically authorized.

Validate activity against expected supplier behavior and subsequent resource operations.

Investigate Authority Created After Authentication

The next step is to search for unauthorized modifications that could survive revocation of the original supplier credential.

Relevant changes include application credentials, additional service principals, application owners, consent grants, app-role assignments, directory roles, user accounts, authentication methods, and Conditional Access policies.

Follow suspicious changes into the identities and resources they affect.

For example, an unfamiliar application credential added during the compromise window should be investigated against its initiating identity, approval record, and subsequent use.

If it was created without authorization, removing the original ConnectWise credential does not address that second credential.

The same principle applies to new privileged roles, mailbox delegation, forwarding rules, and additional application access.

Establish the Downstream Resource Outcome

Authentication logs answer who or what authenticated, where, and when.

They do not necessarily answer which sensitive data was accessed or modified afterward.

Review available Microsoft Graph activity, Microsoft 365 audit records, and relevant Azure resource telemetry.

Where the affected identity could reach Azure Key Vault, inspect applicable data-plane records for secret access and key operations.

Where the identity had Microsoft 365 data permissions, examine relevant mailbox, SharePoint, OneDrive, and other resource activity.

Do not infer that the absence of Microsoft Graph activity logs proves no Graph calls occurred. Microsoft documents that Graph activity collection must have been enabled and routed to supported storage or analytics destinations.

A current configuration change cannot reconstruct previously uncollected operations.

A Tenant-Level Investigation and Remediation Worksheet

The following worksheet converts the trust relationship into named responsibilities and measurable exit conditions.

ActionAccountable ownerRequired evidenceDependencyCompletion condition
Identify affected applicationsCustomer identity ownerTenant ID, application IDs, local object IDs, current and historical integration recordsSupplier application inventoryRelevant applications have recorded dispositions
Preserve available evidenceSecurity investigation leadEntra logs, Microsoft 365 audit records, available Graph and resource telemetryLog access and retention coverageEvidence is protected and missing periods are documented
Restrict affected trustIdentity administratorLocal application state, supplier guidance, enforcement observationsTrusted administration and service-impact assessmentAffected access is denied within the stated scope
Investigate authenticationSecurity operationsSuccessful and failed service-principal sign-ins, source addresses, resource identifiersKnown expected activitySuspicious authentication has a documented disposition
Investigate persistenceIdentity and application ownersCredential, owner, consent, application-role, and privileged-role changesConfiguration history and approvalsConfirmed unauthorized access has been removed
Investigate resource effectsData, application, and security ownersGraph, Microsoft 365, Key Vault, and other applicable resource recordsResource-specific audit coverageConfirmed impact and unresolved exposure are distinguishable
Assess backup continuityBackup service ownerLast successful snapshot, interrupted interval, restore availabilitySupplier service stateProtection gap and business response are recorded
Reauthorize replacement serviceTenant administrator and supplier-service ownerApproved new application, consent permissions, authentication stateSecurity review and vendor restoration instructionsOnly approved replacement authority is enabled
Validate recoveryBusiness-service ownerNew successful snapshot, restore validation, monitoring, and security acceptanceCompleted remediation and reauthorizationService operates within an accepted trust boundary

This worksheet should be completed independently for the managed service provider’s own tenant and for each customer tenant.

A single partner-level finding does not establish the state of every downstream customer.

Restoring Cloud Backup Without Reintroducing the Incident

The ConnectWise restoration sequence illustrates an operational tradeoff.

The supplier needed to protect customers by disabling the affected connections. That containment also interrupted scheduled backups, backup archives, and restore access during the initial service suspension.

The resulting recovery problem had two dimensions.

The first was security: remove the compromised access and investigate customer-tenant persistence.

The second was service continuity: restore backup and recovery capabilities without reestablishing the same vulnerable trust relationship.

What ConnectWise Reported by November 4

ConnectWise’s November 4 update stated that Microsoft 365 backup and restore capabilities were available for reauthorized customer tenants.

It also reported that the end-customer portal and backup-monitoring dashboards were available. Backup Health reporting, export jobs, and Marketplace APIs had been restored, while some billing, notification, and platform dashboard capabilities remained incomplete.

The supplier reported that existing backup archives remained encrypted, isolated, and preserved, with no indication that those archives had been compromised.

That is a vendor statement about the retained backup archives. It is not a forensic conclusion about every customer’s live Microsoft 365 environment.

ConnectWise also described moving the affected application connections toward federated identity credentials as part of its hardening work.

Federation can reduce reliance on stored reusable application secrets, but it does not remove the need to govern the trusted issuer, subject, audience, permissions, and application lifecycle.

Reauthorization Is an Administrative Decision

ConnectWise began restoration through a new Cloud Backup for M365 application.

Partners and customer administrators were instructed to reauthorize tenants using the new connection and to leave previously disabled applications disabled unless specifically directed otherwise.

That matters because the reauthorization operation creates or renews an application trust relationship.

Before consenting, the responsible administrator should verify that the request belongs to the approved replacement application and review the permissions displayed during the consent process.

The business service owner should also understand what backup capability depends on that grant and whether the access is appropriately scoped.

Do not restore the prior application simply because it appears familiar or because a backup job is waiting.

A Successful Reconnection Is Not a Successful Backup

ConnectWise explicitly cautioned that reauthorization alone does not prove backups have resumed.

The customer or partner must verify that new backup snapshots complete successfully for each relevant tenant.

Snapshot history may contain a gap between the last completed backup before the interruption and the first successful backup afterward.

That gap is an operational fact to investigate, not merely a cosmetic reporting issue.

Data may have changed during the interruption. The current protection point may therefore differ from the expected recovery point.

The backup owner should record the last pre-interruption successful snapshot, the first post-restoration successful snapshot, the service’s recovery objectives, and any business impact from the gap.

Where appropriate and supported, validate representative restoration rather than relying solely on job-completion status.

An encrypted archive remaining intact is valuable. It does not establish that recently changed customer data was protected during the outage.

Designing Better Supplier-Integration Boundaries

The ConnectWise case is a reason to redesign the review of third-party application authority.

The objective is not to prohibit multitenant applications. Many important enterprise services rely on them.

The objective is to ensure that the permissions, credentials, monitoring, recovery procedures, and supplier responsibilities reflect the actual consequence of compromise.

A Supplier-Integration Review Worksheet

Review dimensionArchitecture questionEvidence to retain
Business purposeWhich service outcome requires this integration?Service description, owner, dependency map
Application identityWhich registration and local service principals represent the integration?Application ID, tenant object IDs, publisher and owners
Effective permissionsWhat data and administrative actions are actually reachable?Granted roles, consent records, resource scopes
Credential managementWho issues, stores, rotates, and revokes authentication credentials?Credential lifecycle and supplier responsibilities
Credential expansionCan this identity add credentials or privileges to another identity?Effective authorization and change history
MonitoringCan the customer distinguish routine supplier activity from anomalous use?Service-principal logs, resource activity, alert ownership
Supplier incident coordinationHow will the customer receive indicators, scope updates, and restoration instructions?Contact and notification process
Emergency containmentCan access be restricted without depending on the compromised supplier channel?Authorized procedure and tested administrative access
Service fallbackWhat happens when the integration must remain disabled?Backup, recovery, and continuity alternatives
Return to serviceWhat must be demonstrated before consent or authority is restored?Security gates, functional checks, and service-owner acceptance

This worksheet is intended for architecture review and service onboarding, not as an assertion that every vendor supports every ideal control.

Some suppliers may require broad permissions because of how their supported APIs operate. When those permissions cannot be reduced, the organization should address the resulting exposure through stronger monitoring, emergency restriction procedures, supplier requirements, and tested recovery alternatives.

The exception should be explicit and owned.

Limit Cross-Tenant Administrative Reach

An MSP may manage hundreds of customer tenants, but its internal operating convenience should not automatically create one undifferentiated administrative boundary.

Assess whether credentials and roles are shared across tenants, whether customer administration can be constrained independently, and whether support workflows require more authority than the service normally needs.

Use least-privilege application grants and separate high-impact administration where feasible.

For customer-owned configurations, maintain records of which applications are authorized and who can modify that authorization.

An enterprise should be able to answer which supplier identities have access before an incident requires urgent revocation.

Make Customer Telemetry Part of the Agreement

The supplier may not have visibility into every customer’s audit logs, and the customer may not have visibility into the supplier’s credential-management environment.

That is a real accountability boundary.

The contract and operating model should identify which party preserves authentication history, resource audit records, application changes, and incident indicators.

They should also define how relevant information will be exchanged during an investigation.

Neither party should assume the other possesses the complete evidence chain.

Design for Supplier Unavailability

Supplier compromise can require service suspension even when no customer data loss has been established.

For a backup service, that creates a particularly difficult situation: the organization may lose some of its normal protection and recovery capability while responding to a security incident.

The architecture should therefore consider how critical data is protected if the supplier connection must remain unavailable.

An alternate recovery path can reduce dependency risk, but it also introduces cost, data consistency, operational complexity, and new administrative responsibilities.

Those tradeoffs belong in the service’s resilience design rather than being discovered when emergency containment interrupts normal backups.

What This Incident Does Not Prove

Several boundaries matter for accurate reporting.

The investigation does not establish a vulnerability in Microsoft Entra or Microsoft Graph. Microsoft explicitly stated that the ARTEX-associated activity did not result from compromise of Microsoft corporate systems, products, or services.

The attempted authentication of more than 2,250 customer tenants does not establish an equal number of successful intrusions or data-theft victims.

ConnectWise’s findings about encrypted backup archives do not independently establish whether unauthorized activity occurred inside each customer’s live tenant.

Okta’s broader identity reconnaissance investigation does not establish that every related operation was conducted by Storm-0940, nor that every observed framework instance was ARTEX.

The Cloud Backup incident also should not be merged with later ScreenConnect and SCKit reporting without evidence of a shared technical intrusion.

These distinctions directly affect remediation. They prevent organizations from treating unrelated systems as confirmed compromises while overlooking the specific permissions, credentials, and resource operations that are supported by the evidence.

Practical Acceptance Gates for Restoring Supplier Trust

Before returning the integration to normal operation, require four separate outcomes.

The affected trust is restricted. Vendor-side compromised credentials have been revoked, the appropriate local enterprise applications have been reviewed, and relevant access restrictions are demonstrably effective.

The customer environment has been investigated. Available authentication and resource records have been reviewed, unauthorized follow-on changes have been addressed, and missing evidence is recorded as an investigation limitation.

The replacement trust has been approved. The new application identity, requested permissions, credential-management model, and responsible owners are understood and accepted.

The business service is demonstrably working. Backup snapshots and required restore functions have been validated, monitoring is operating, and any protection gap or remaining service limitation has an accountable owner.

These gates should be recorded separately.

An administrator successfully consenting to an application does not satisfy the security investigation. A security team finding no recent suspicious sign-ins does not prove that an older unlogged operation never occurred. A vendor declaring service available does not prove that every customer’s backups resumed.

The service should return to a bounded operating state supported by evidence, with explicit residual risks and a tested means of restricting access again if necessary.

Conclusion

The Storm-0940 and ConnectWise investigation demonstrates how a trusted supplier application can become an access path into customer environments without requiring a new vulnerability in the customer’s identity platform.

ARTEX accelerated attacker activity, but the fundamental security issue was the authority attached to established application trust.

Microsoft documented attempted authentication across more than 2,250 tenants and confirmed follow-on activity in some environments. ConnectWise revoked affected vendor credentials and restored services through a new application. Customer tenants still required independent review because vendor-side containment did not automatically reverse unauthorized changes inside those tenants.

The lasting architecture lesson is to treat supplier applications as consequential nonhuman identities. Their permissions, credentials, reachable resources, incident controls, and recovery dependencies should be reviewed with the same seriousness as privileged human access.

The operating question is:

If a supplier’s application credentials were compromised, could we identify every tenant it reaches, determine what changed, remove unauthorized access independently, and restore the service without reopening that trust boundary?

That is the standard a mature supplier integration should be designed to meet.

For the mechanisms that can survive the original credential’s revocation, read the persistent-access eradication guide. Microsoft’s compromised-application investigation playbook provides the platform-specific investigation and containment workflow.

External References

The post Storm-0940 and ConnectWise: How Stolen Application Trust Reaches Customers appeared first on Digital Thought Disruption.