Deception Check

SKYKICK / MICROSOFT 365 RESPONSE

Disable the connection.
Follow the access.

The SkyKick incident calls for more than checking the backup application. Review access added to other identities, then account for the recovery gap.

A vendor can revoke its credential. Your tenant can still need investigation.

That is the response problem in the ConnectWise Cloud Backup incident, involving the service formerly known as SkyKick. The useful question is not just whether the original connection is disabled. It is whether someone used that access to establish another way in, and whether your team has the records to find it.

Detection resources updated October 29, 2026. Native-source hunts now lead the companion; the incident snapshot remains dated October 19.

Two observations that change the response

Huntress reports unauthorized Cloud Backup application sign-ins. In its original persistence case, a Cloud Management application for Microsoft Teams assigned itself Global Administrator and added client secrets to two other trusted applications. Huntress later reported use of one planted credential.

October 19 update: Huntress identified a second tenant with the same role-assignment and credential-planting pattern. The original case illustrated below remains useful: disabling the original application does not remove credentials already added elsewhere.

Two separate reported observations. Cloud Backup credential abuse led to app-only tokens and access bounded by tenant grants. In Huntress's original one-tenant persistence case, a Cloud Management MS Teams application added Global Administrator and credentials to two other applications. No arrow joins the observations into one proven timeline.
Huntress's original reported case, simplified. The lower lane describes the first tenant; its October 19 update reports a second tenant with the same pattern. These are third-party findings, not Deception Check fleet captures. The diagram does not join the earlier persistence activity and later sign-in sweep into one proven timeline.

The credential and the permission live in different places

For a multitenant application, the vendor manages its application registration. Each customer tenant has a local representation, the service principal, with that tenant's permissions and configuration. An application's display name is a label; its Application ID and local Object ID identify the objects you need to investigate. Microsoft explains this relationship here.

App-only authentication does not ask a human to complete MFA. That is different from defeating a user's MFA challenge. Nor can you assume a workload-identity Conditional Access policy covers this connection: Microsoft excludes third-party SaaS and multitenant applications from that feature's scope. Keep user MFA; investigate the application identity on its own terms. Microsoft workload-identity policy scope.

Start with the tenant, not the product name

Use ConnectWise's investigation guide for the current response steps. Verify the local enterprise-application settings in your MSP tenant and each customer tenant, including dormant connections. Keep affected connections disabled pending formal restoration instructions. Preserve evidence without delaying urgent containment.

Seven Application IDs in the vendor guidance

Record the local Object ID too. Vendor-owned registrations generally will not appear in a customer's App registrations list. Inspect the local enterprise application instead.

ConnectWise applicationApplication client ID
Cloud Backup506e08f9-6e9a-4da4-942b-a099a03644ae
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

Source: ConnectWise technical guide, October 18, pages 2-3. Other similarly named applications require an identity and permissions review, not guilt by name.

As a working habit, give every tenant a named owner and a response record. A reseller's incident email is not evidence that the customer's administrator has completed a review. Track receipt, ownership, findings, and the next decision separately.

Follow the access through three evidence layers

01 / AUTHENTICATION

Which identity got in?

Review service-principal sign-ins, including successes and failures. Capture the app, tenant, resource, address, result, and credential identifiers where present.

A successful sign-in is not a record of every subsequent action.

02 / AUTHORITY

What access changed?

Follow directory audit events to the target objects: credentials, roles, owners, and consent. Include management and security applications outside the vendor's app list.

An audit event and a credential's current configuration answer different questions.

03 / RESOURCE ACTIVITY

What was done with it?

Correlate available Graph activity and Microsoft 365 audit records with the suspect identity and time window. Review mailbox rules, forwarding, delegation, and file activity.

Record the actual logging coverage before concluding that a search was clean.

Technical references: service-principal sign-in fields, directory audit fields, Graph activity logging, and Microsoft 365 audit activities.

When an audit event points to an added credential, inspect the target application or service principal's credential metadata. Record the credential key ID or certificate thumbprint, its target identity, and the approved change record, if any. Then look for matching authentication records and check their outcomes: a failed attempt is not successful access. No match cannot establish that an issued token was never used. An authorized responder should remove confirmed unauthorized access and assess related identities. Microsoft's compromised-application playbook provides the broader investigation and recovery framework.

Make a clean search explain itself

Before running a hunt, record the tenant, data source, earliest and latest available records, export time, and known gaps. Entra's standard retention is seven days on Free and 30 days on P1/P2 for audit and sign-in logs; separately retained exports may extend coverage. An upgrade does not bring expired history back. Microsoft retention reference.

Records available

The source contains events covering the required window. Check for gaps within it.

No matching events

The tested query found none in those available records. Record the query and its scope.

Not collected or expired

This is a coverage gap. Keep the unanswered question in the incident record.

This distinction is particularly important for Graph requests. Graph activity logging needs to be configured; enabling it now is useful for future visibility, not a reconstruction of last week's traffic. Entra directory audit logs are not a complete ledger of everything an app read. Graph activity logging overview.

USE IT IN YOUR RESPONSE

The cross-platform detection and response companion

The refreshed companion starts with documented Microsoft cloud records, Sophos query sources, and raw Microsoft Graph directoryAudit JSON searches for Splunk, LogScale, and Sumo Logic. It includes two standard Azure audit-log Sigma starters. Choose your source and explore the queries before interpreting a result. The earlier dc_* normalized examples remain in a separate advanced appendix for teams building integrations.

Download the full companion (PDF) Download the PDF and code bundle (ZIP)

The tenant worksheet turns the investigation into an accountable handoff. The focused KQL starters cover record availability, application sign-ins, directory changes, and a credential-use pivot.

Tenant response worksheet (PDF) · KQL starters · Query guide and KQL interpretation

Carousel text transcript · Six-card carousel (PDF)

Read-only investigation starters, not deployed detections. Parser, schema, and synthetic checks are documented separately for each query family; no customer-tenant or service runtime testing has been performed. Native-source routes have source and local fixture checks, not customer-tenant execution. Test against known events before relying on an empty result.

Track the recovery gap alongside the security work

Give the business a separate answer about recoverability. For each critical workload, record its last successful recovery point, the age of that point, whether a restore is currently available, and who owns the interim continuity plan. Avoid a single green "backup protected" status that hides unavailable restores or a growing gap between copies.

Restoration should have its own gate: formal vendor instructions, authorized tenant changes, completion of required credential and access cleanup, then verification that connections and backup jobs work. A successful new job is useful evidence. It does not settle the earlier intrusion question.

The question to carry into the next vendor review

If this application's credential were abused tomorrow, could we identify what it changed, what it accessed, and which access survived containment?

Ask that during integration reviews, before the incident call. Test the handoff from a suspicious application event to a responder who can inspect the target object and preserve the evidence. Our goal is an early warning your team can investigate, not another alert with no next step.