If your organisation uses ConnectWise Cloud Backup, formerly SkyKick, ask the administrator responsible for Microsoft 365 to review its application access and your tenant’s activity. Restoring the provider’s service and establishing that your own environment is safe are separate tasks.
Sources checked on 22 October 2026. This is an evidence-based guide to the incident and its operational consequences, not a report that your organisation has been compromised or that I have investigated a customer tenant.
The difficult part of a cloud backup incident is that the backup service can have permission to work with your live information. The archive is one asset; the identity used to read or manage Microsoft 365 is another. A statement about the condition of stored backups does not answer what that identity did inside a customer environment.
For a business owner, the first useful question is therefore not just “Are our backups intact?” Ask which connected applications had access, who is reviewing their activity, what happened during the interruption and what evidence will support reconnection.
What is confirmed as of 22 October
ConnectWise’s advisory update, dated 22 October at 00:45 UTC, describes unauthorised access to Cloud Backup application credentials. Phased restoration is underway. The vendor says existing archives are encrypted, isolated and preserved, with no indication of compromise, while individual Microsoft 365 tenants still require review. Keep affected connections disabled until formal instructions apply to your service. Current ConnectWise advisory.
Treat that as a dated service update. It is neither a guarantee about every customer tenant nor a promise that every backup, restore and connected function is available. Before making a decision, open the current advisory and record which update you relied on.
Microsoft’s 21 October report, following the vendor’s initial 16 October advisory, assesses that Storm-0940, a China-state-aligned actor using the ARTEX AI framework, was responsible for the campaign. It reports attempted sign-ins to more than 2,250 tenants, not 2,250 confirmed breaches. Its separate 10–14 October password-spray sample targeted 327 accounts across eight organisations without observed successful authentication. Other app-credential activity did produce authentication and subsequent cloud changes in multiple environments. Microsoft investigation.
Those distinctions matter. A failed sign-in, a successful application authentication, a new privileged assignment and a downloaded business document are different observations. An incident review should explain which of them is supported by the records for your tenant.
Does this concern my organisation?
Your organisation may buy backup directly, through an IT provider or as a line inside a wider managed-service agreement. The name on the monthly invoice may not reveal which application identities remain authorised in Microsoft 365.
| Your situation | What to establish | Who should own the answer |
|---|---|---|
| You use the affected Cloud Backup service | The connected applications, permissions, service state and tenant review | Microsoft 365 administrator and backup provider |
| An MSP manages Microsoft 365 for you | Whether your tenant was connected, including older SkyKick arrangements | MSP and your named internal service owner |
| You stopped using the service previously | Whether application access or credentials remain, and what historical evidence is available | Tenant administrator |
| You use another ConnectWise product | The exact product and integration; a brand name alone does not establish scope | Product owner and supplier |
| You do not know which backup system is used | The backup contract, administrative owner and actual tenant connections | Business owner and IT provider |
Do not ask an ordinary employee to explore unfamiliar admin settings or delete applications. Assign the review to someone who can inspect the right environment and understands the effect on backup, security monitoring and other connected work.
An MSP also needs a tenant-by-tenant list. A completed review of the provider’s own Microsoft 365 environment does not establish the state of each managed customer. Record tenants with no match, tenants with a match, tenants requiring investigation and tenants for which evidence is incomplete.
Why an application can access data without your password
A background integration needs to work when no employee is sitting at a browser. Microsoft supports application-only access, in which software authenticates with its own identity and credentials. That is distinct from delegated access, where an application acts for a signed-in user. The permissions granted to the application define its reach. Microsoft Graph permission models.
A simplified example is a backup integration authorised to read a workload. It obtains an access token using its application identity, then calls the relevant service. If that application credential is misused, the attacker is not necessarily entering an employee’s password or completing an employee’s sign-in prompt. The client-credentials flow is specifically designed for software operating without an interactive user. Microsoft’s client-credentials documentation.
This is why changing a user password alone is not an adequate response to compromised application credentials. It addresses a different identity. Multi-factor authentication remains important for people; this case does not establish that MFA stopped working. The responder must identify the application’s credential and permission path and contain that path appropriately.
Think of the distinction as two separate access records in your service inventory. One describes who may sign in as a person. The other describes which software may act in the background, with what authority, for whose business purpose and under whose ownership.
Application ID and service principal ID are different identifiers
An application has an application or client ID. A customer tenant also has a local representation of that application, called a service principal, with its own object ID. A multitenant application can therefore appear in several organisations without their local service-principal object IDs being identical. Microsoft’s explanation of application and service-principal objects.
| Field to record | Why it matters |
|---|---|
| Customer tenant ID | Identifies the organisation whose records are being reviewed |
| Application/client ID | Identifies the vendor application across tenants |
| Local service-principal object ID | Identifies its representation inside this tenant |
| Display name and publisher | Useful context, but not a substitute for stable IDs |
| Permissions, roles and owners | Defines the access being investigated |
| Credential identifiers and change times | Supports review of authentication and configuration changes |
Use the vendor’s complete current application list, including legacy and precautionary entries. Do not treat a short sample query from an investigation article as an exhaustive inventory. ConnectWise application guidance and FAQs.
For each match, record the identifiers before changing configuration. Look for an old integration even when nobody recognises its present display name. If an application has already been removed, its absence today does not recreate its earlier permissions or explain its historical use.
Independent investigations show why local evidence matters
Huntress reported activity on 15 October in two tenants where the application assigned itself Global Administrator and credentials were added to other locally registered applications. That illustrates how access can outlast the original integration. It is a report about observed cases, not a declaration that every connected tenant experienced those changes. Huntress investigation.
Sophos described app-only token activity in environments it reviewed without finding the same additional persistence there. Its observations have their own scope. They do not invalidate a different responder’s findings in other environments. Sophos investigation.
Todyl also documented automated application access, including an example involving a subscription that was no longer active. A cancelled commercial service and a removed technical permission are different states. Todyl investigation.
Your administrator’s report should use the same discipline: say which tenant, time range, identity and record support the conclusion. “Other customers were unaffected” is not evidence about your organisation. Nor should evidence from a small number of compromised tenants be expanded into an unsupported claim about every customer.
Preserve evidence before making irreversible changes
An application incident can require urgent containment and careful preservation at the same time. Capture the relevant configuration and available records before deleting objects or removing the context needed to interpret activity. Do not delay necessary containment while waiting for a perfect export. Microsoft’s incident-response playbook treats investigation, scoping and remediation as related stages. Compromised-application response playbook.
Agree who may make changes and where the evidence will be stored. Exports can contain personal information, filenames, account details and sensitive service configuration. Use an access-controlled case location rather than sending raw records through a general support mailbox or pasting them into a public forum.
A practical evidence register should identify the source, tenant, time zone, covered dates, export time and person responsible. Keep the original export and a separate working copy for analysis. If the review crosses several providers, make the handover explicit so each team knows which records it is expected to preserve.
A useful change record answers four questions: what was changed, who authorised it, why it was necessary and how the operational effect was checked. That record helps distinguish a responder’s own credential rotation from an earlier unexplained credential addition.
Look at application sign-ins, not only employee logins
Microsoft Entra provides service-principal sign-in records for application authentication. A review limited to the normal user sign-in screen can miss the identity involved in this kind of event. Examine the application, local service principal, result, source, resource and timing together. Service-principal sign-in documentation.
A success result needs interpretation against expected vendor activity. A failure establishes an unsuccessful request, not access to a mailbox. An unusual source or a burst of activity is a reason to investigate; it is not by itself a final finding about business-data theft.
The following is an original, read-only investigation example for a Log Analytics workspace that already receives the relevant table. Replace the placeholder with a verified application ID from the current vendor guidance. Repeat for the other applicable IDs, select an appropriate investigation period and validate the available schema with your administrator. It has not been run against a customer environment for this article.
AADServicePrincipalSignInLogs
| where TimeGenerated >= datetime(2026-10-01)
| where AppId == "<verified-application-id>"
| project TimeGenerated, AppId, ServicePrincipalId, IPAddress,
ResourceDisplayName, ResultType, CorrelationId
| order by TimeGenerated asc
The table, fields and their meanings are documented by Microsoft. This example does not search every log source, prove that telemetry was collected or perform remediation. Do not add a narrow list of published IP addresses as the only condition and then interpret no results as an all-clear. AADServicePrincipalSignInLogs reference.
Connect authentication with what happened afterwards
Successful authentication establishes that an identity authenticated, not everything it subsequently did. Review directory changes and relevant workload activity alongside sign-ins. Ask whether credentials, owners, permissions, roles or other identities changed without an authorised business reason.
A credential added to another application deserves separate attention: disabling the original integration may leave that different access route available. The response must follow the changes discovered in the tenant, rather than end after one vendor connection has been disabled.
Microsoft’s application security-operations guidance identifies credential changes, privileged permissions and unusual application activity as monitoring concerns. Use those categories to structure review, while checking each finding against legitimate administration and deployment work. Microsoft Entra application security operations.
Where Microsoft Graph activity logs exist, they can add request-level context about resource access and API operations. They require the appropriate licensing and prior logging configuration; enabling them today does not produce a historical record of yesterday’s requests. Microsoft Graph activity-log documentation.
For email and document workloads, ask what audit records are actually available for the relevant operation. A response saying “we searched the audit log” should identify the service, dates and events searched. Microsoft’s Purview audit-search guidance explains the separate workload-audit search process. Microsoft 365 audit search.
Missing logs are a finding about visibility
Default Microsoft Entra retention varies with licensing: the documented baseline is seven days for Free and 30 days for P1/P2 sign-in and audit records. Other destinations or retention arrangements may preserve more. Upgrading a licence cannot bring back records that have already expired. Entra retention documentation.
Keep an explicit list of evidence gaps. Perhaps the application existed during the period, but the relevant sign-in records are no longer retained. Perhaps authentication is visible while Graph activity was not collected. Perhaps the provider can supply records that your organisation cannot access directly.
These situations call for a qualified conclusion, not a confident “nothing happened.” For example: “We found no unauthorised activity in the retained records from these dates; earlier activity could not be assessed from this source.” That statement is more useful than a clean-sounding result whose time boundary is hidden.
If an analyst cannot obtain a particular record, document who was asked and what alternative evidence was reviewed. Do not fill the gap with a guessed victim count or assume a service-status page is a customer-specific forensic report.
Contain the access path without destroying the investigation
Use the current vendor instructions and the tenant administrator’s response plan to disable affected access, preserve its configuration and review the consequences for connected work. Do not delete every application carrying the vendor’s name or remove unrelated identities as a shortcut. The vendor’s technical guide is the incident-specific reference. ConnectWise technical investigation and remediation guide.
Containment should address both the original route and any unauthorised changes actually found. Have an authorised responder review suspicious credentials, grants, owners, roles and accounts before removing or replacing them. Record the intended post-change state and verify it afterwards.
For a customer using an MSP, agree who controls the local tenant object and who controls the vendor-held credentials. A customer administrator may be able to disable access locally without being able to rotate a publisher’s secret. Avoid a response plan in which both parties assume the other has completed that work.
User-password resets can still be appropriate where user-account exposure is found. They should be part of the scoped response, not presented as a universal way to revoke application-only access. Similarly, retain MFA and improve administrative controls without claiming that an ordinary user MFA policy covers every workload identity.
Protect backup continuity during the interruption
Security containment can create an availability problem even when stored archives have not been reported as compromised. Establish when the last successful backup completed, which workloads it covered and which restore paths can actually be used today.
Do not infer a successful backup from a scheduled job or an old green dashboard. Ask for the completion record and the recovery point. If the interruption spans several working days, explain the resulting gap in terms the business understands: which newly created or changed information would need another recovery route.
The NCSC’s cloud-backup principles emphasise protected recovery access and restoration testing. Verify recovery access and dependencies before an emergency makes the backup system itself unavailable. A second copy should have a defined purpose, protection and ownership; merely buying another product does not establish that it can restore the required data. NCSC cloud-backup principles.
For example, an organisation may need a recently approved contract, a shared project folder and a mailbox to resume work. List the recovery point required for each, the system that holds it and the person authorised to restore it. A website database backup will not automatically cover Microsoft 365 email, and a cloud-mail backup will not automatically cover a separate web server. NCSC technical recovery guidance.
If an alternative backup or export is introduced during the incident, assess its permissions and storage before connecting it. An urgent workaround should not create a second unreviewed integration with broad access and no owner.
Reconnect only through a controlled restoration plan
Restoration has at least three workstreams: the supplier restores its service; the customer reviews the tenant; and the business confirms that protection and normal work have resumed. Those workstreams can progress at different speeds.
Before authorising reconnection, ask for the vendor’s applicable instructions, a named tenant reviewer and a record of remaining security questions. Identify which identity will be connected and what permission changes are expected. Compare the resulting access with the intended service purpose.
Microsoft recommends granting applications only the permissions their work requires. That principle is useful during restoration: understand every grant you approve, and distinguish a necessary service permission from an unexplained privileged role. Some backup tasks legitimately require substantial access; document the reason rather than assuming every broad permission is either automatically malicious or automatically acceptable. Least-privilege application guidance.
| Restoration question | Evidence to retain |
|---|---|
| Is this service ready to reconnect? | The applicable vendor instruction and its date |
| Has our tenant been reviewed? | Scope, findings, remediation and unresolved evidence gaps |
| Is the intended application connected? | Verified application and local service-principal IDs |
| Are permissions appropriate? | Reviewed grants, roles, owners and business purpose |
| Are backups working again? | A completed job, current recovery point and agreed restore check |
| Who will watch for recurrence? | Monitoring owner, review period and escalation route |
Restoring a login page is not the same as restoring a reliable backup process. Verify an actual job and an appropriate recovery test through the responsible administrator, while protecting production data and avoiding an uncontrolled overwrite of current business information.
What the AI aspect changes—and what it does not establish
The reported use of ARTEX is relevant because automation can compress the time available for response. It does not change the need to identify permissions, preserve records and close persistent access. Nor does it establish that an AI system independently discovered the vendor’s initial entry route.
Keep attribution and impact separate in customer communications. The business needs to know which information and services may be affected, what is being done and when the next evidence-based update will arrive. Naming an attacker or an AI tool does not complete those tasks.
Avoid describing the published sign-in-attempt count as a confirmed number of breached organisations. Also avoid generalising findings from one reviewed tenant to every tenant on the platform. A useful report can be serious without turning uncertainty into a larger headline.
A request to send your IT provider
The following is an original request you can adapt. It is not a declaration that your tenant has been compromised:
Please confirm whether our Microsoft 365 tenant currently has, or previously had, any application connections covered by the ConnectWise Cloud Backup incident guidance, including legacy SkyKick arrangements. Identify the relevant application and local service-principal IDs, the available investigation period and the records reviewed. Explain the containment and tenant-specific findings, any unauthorised changes remediated, remaining evidence gaps and the vendor instructions required before reconnection. Also confirm our last successful backup, the current recovery gap and how normal backup and restore capability will be verified. Please distinguish completed work from open questions and identify the responsible owner for each.
Ask for a report for your own tenant, not just a copy of the vendor’s public announcement. If several suppliers are involved, make sure the report identifies who is coordinating the response and who can authorise each change.
For a small organisation, a concise written record can be enough to make ownership clear. It should still name the environment, the evidence period, what was observed, what changed and what remains uncertain. Preserve that record with the maintenance and recovery documentation.
Keep the website and its connected services in the same maintenance map
Your website, Microsoft 365 tenant, backup provider and deployment systems can be run by different people. Put their dependencies in one service map so an incident in one system does not leave the others without an owner or a recovery route.
The Rank Math PRO and Node.js guide covers a different combination of malicious packages and runtime flaws. The PHP security article concerns server-runtime patches. None of those updates replaces the tenant-specific identity review described here.
For help clarifying the maintenance of an existing website and its integrations, describe your website and its current support arrangement. Microsoft 365 incident investigation and changes should remain with the authorised tenant administrator and security provider. Share sensitive records only through the agreed secure response process.