Illustrative backup monitoring desk with blue status displays, an amber warning beacon, and an empty operator chair
Cloud Security October 30, 2026 Huntress / ConnectWise / Microsoft Threat Intelligence / The Record

One Vendor Credential Opened 162 Microsoft 365 Tenants. Backups Stayed Off for 11 Days.

A backup vendor's app was already trusted inside customer Microsoft 365 tenants. When its credential was turned against them, the app could raise its own privileges, and the safest way to stop it also stopped the backups.

By the AuthorityGate Architect Team

In this issue: 4 parts
  1. Part 1. What Happened
  2. Part 2. How a Trusted App Became the Attack Path
  3. Part 3. The Recovery Gap
  4. Part 4. What You Can Do Now

Key takeaways

What happened in the ConnectWise Microsoft 365 breach?

A compromised ConnectWise Cloud Backup credential allowed unauthorized sign-ins to customer Microsoft 365 tenants. Huntress confirmed 162 tenants by October 16, 2026, then reported over 300 with successful sign-ins. Those are observer counts, not a final total. The credential lived in the vendor tenant, outside customer-side rotation control.

Why did ConnectWise backups stop for roughly 11 days?

ConnectWise disabled Cloud Backup application access on October 17 to contain the incident, interrupting automated backups. Phased restoration through a new application began October 28, with each tenant requiring re-authorization. The gap shows why containment and recovery need separate acceptance checks: a restored connection alone does not prove fresh backups or working restores.

What should a validation layer check before backup trust resumes?

Review application roles, credentials, and persistence; route privilege changes to a named human approver. After re-authorization, verify new snapshots and a test restore for each tenant. AuthorityGate Keystone is designed around change validation, known-good evidence, and recovery readiness. This is a governance application of that model, not proof it prevented this incident.

Key facts: citable summary

  • Huntress confirmed unauthorized sign-ins in 162 tenants as of October 16, 19:00 UTC, and later reported over 300 tenants with successful sign-ins. No final compromised-tenant total was published in the cited record. Source: Huntress, October 16 analysis and updates.
  • Cloud Backup application access was shut off October 17; phased Microsoft 365 restoration through a new application began October 28. Existing archives were reported encrypted, isolated, and preserved; the credential root cause remained under investigation. Source: ConnectWise advisory, October 16-November 4.
  • Microsoft attributed the campaign to Storm-0940 and reported use of the ARTEX agentic AI framework. That attribution is a Microsoft assessment, not an independent finding by AuthorityGate. Source: Microsoft Threat Intelligence, October 21.

AuthorityGate analysis: vendor consent must not silently authorize privilege expansion, and containing a backup supplier must trigger independent recovery validation.

Keystone is designed to route changes through dependency checks, behavioral validation, recovery readiness, and named human approval. Detected drift is surfaced for validation and recovery decisions under customer policy; these are early-access design statements. Read the Keystone validation model.

Part 1 of 4

What Happened

In mid-October 2026, an attacker used a Microsoft 365 credential that belonged to ConnectWise Cloud Backup, the service formerly known as SkyKick, to sign in to customer tenants. The backup app had been consented into those tenants so it could read and protect mail, files, and Teams data. That consent did not care who was holding the credential.

Two things make this incident worth studying beyond the security headlines. In at least one tenant, a vendor application raised its own privileges to Global Administrator, and nothing on the customer side asked a person before that change took effect. Then, when ConnectWise contained the incident, it had to switch off the very connections that kept customer backups running. For roughly 11 days, the organizations relying on that service had no new Microsoft 365 backups from it.

A Credential Customers Could Not Rotate

Huntress, a managed security provider, published its analysis on October 16. Its summary was direct: "A Microsoft 365 credential (API secret) belonging to ConnectWise Cloud Backup (formerly SkyKick) has been compromised." The sign-ins came from a hosting provider address, 144.172.64[.]128, and received Microsoft Graph access tokens. Huntress noted that legitimate traffic for this app comes from Azure private IP space.

The credential type matters. Huntress described it as a client assertion, "which means the credential lives in the vendor's own tenant. Nothing a customer did in their environment could have produced it." It also means nothing a customer could do inside their own tenant would rotate it. The credential "works against every tenant the app is consented in," Huntress wrote, "so the actor is not limited to the tenants we have confirmed so far."

The counts grew as visibility grew. As of 19:00 UTC on October 16, Huntress had confirmed unauthorized sign-ins in 162 tenants. A later update reported almost 1,000 Entra tenants with sign-in activity from the same address, and over 300 of those tenants saw successful sign-ins. These are Huntress's figures from what it could observe, so 162 is best read as a floor, not a total. Microsoft, from its own telemetry, later said the actor attempted to sign in to more than 2,250 tenants. No source has published a final count of compromised tenants.

Containment Switched Off Backups

ConnectWise posted its initial notice at 17:47 UTC on October 16, saying there was "no indication that partner backup data has been accessed." At 05:01 UTC on October 17 it took the decisive step: "we are shutting off Cloud Backup application access. This will interrupt backup services for customers until further notice." By 07:00 UTC that day it had disabled seven ConnectWise application identities. Later that day it revoked certificates and said the shared credentials previously used for Cloud Backup were no longer valid.

ConnectWise reported completing a full reset of all accounts and credentials in the affected Cloud Backup environment on October 19. On October 21, it said that, based on the evidence available to it and threat intelligence shared by Microsoft, it appears a China state-aligned group was responsible. A target of October 26 for partner onboarding slipped, and at 16:00 UTC on October 28, phased restoration began, starting with Microsoft 365 backups and restores through a new, security-hardened application. Partners had to re-authorize each customer tenant.

Recovery of the wider service continued after that. ConnectWise's November 4 update said backup health reporting and export jobs had returned on November 3, while billing functionality was still in progress and the platform backup dashboard was not yet fully restored. Its October 30 update said existing backup archives "remain encrypted, isolated, and preserved, with no indication of compromise." How the attacker obtained the credential "remains under investigation," according to ConnectWise.

Timeline: October 10 to November 4

When (2026, UTC) Event What it tells you
Oct 10-14 Password spraying from the same 144.172.64.x range; Microsoft reports a peak of 2,900 failures per minute on Oct 11 Noisy activity on shared infrastructure came before the app abuse
Oct 15, from 01:58 In roughly 50 Huntress-monitored tenants, ConnectWise Cloud Management service principals assign themselves directory roles Huntress says the user agent matches vendor automation, unconfirmed; either way, no customer approval step
Oct 15, 03:44-03:49 In one tenant, the Teams management app adds itself to Global Administrator and plants client secrets set to expire in 2028 Persistence in about five minutes
Oct 16, 17:47 ConnectWise posts its initial advisory; Huntress publishes its analysis About a day and a half after the role changes began
Oct 16, 19:00 Huntress confirms unauthorized sign-ins in 162 tenants A floor from one observer, not a total
Oct 17, 05:01 ConnectWise shuts off Cloud Backup application access; backups interrupted until further notice Containment and backup coverage are now in direct conflict
Oct 17, 07:00 Seven ConnectWise application identities disabled; certificates revoked later that day Disabling stops new sign-ins, not changes already made
Oct 19 Full reset of accounts and credentials in the affected environment The vendor side is cleaned; customer tenants still need their own review
Oct 21 Microsoft attributes the campaign to Storm-0940 and the ARTEX framework Agentic AI tooling on the attacker side
Oct 21-26 ConnectWise says it has begun restoring services in controlled phases; partner onboarding targeted for Oct 26 Vendor-side restoration and validation, not yet partner backups
Oct 28, 16:00 Phased restoration begins with Microsoft 365 backups through a new application; each tenant must be re-authorized About 11 days without new backups from the service
Nov 4, 18:55 Billing still in progress; platform backup dashboard not yet fully restored Recovery is a sequence, not a switch
Close-up of a dark storage array in a server rack, one drive bay showing a single red warning light
Disabling a compromised vendor app is the right containment step. It can also be the moment your backups stop.

Part 2 of 4

How a Trusted App Became the Attack Path

By the Numbers

162
tenants with unauthorized sign-ins confirmed by Huntress, Oct 16
300+
tenants with successful sign-ins, of almost 1,000 with sign-in activity (Huntress)
7
ConnectWise application identities disabled on Oct 17
<3 hrs
targeting to privileged access in multiple cases, per Microsoft

An App That Granted Itself Global Administrator

The sharpest detail in the Huntress analysis is a five-minute window in one customer tenant. Between 03:44 and 03:49 UTC on October 15, the ConnectWise Cloud Management app for Microsoft Teams added itself to the Global Administrator role, at 03:44 and again at 03:48. In the same window it made nine changes to two other applications, an MSP management application and Huntress's own security application, adding new client secrets. Six of those changes succeeded. The new secrets were set to expire in 2028.

Read that as a change record and the problem is plain. A service identity elevated itself to the most powerful role in the tenant, then created long-lived credentials on other identities. Each of those is a privileged change. In most organizations, a human admin making the same changes would need a ticket, an approver, or at least a second look. The application needed none, because it already held the permission to make them.

The wider pattern is harder to read, and it deserves care. Starting at 01:58 UTC that day, in roughly 50 Huntress-monitored tenants, ConnectWise Cloud Management service principals assigned themselves directory roles, mostly Exchange Administrator. Huntress noted those events carried the user agent ConnectWise's own apps use, which points to vendor automation, but said it had not confirmed that with ConnectWise. In the one tenant where the app made itself Global Administrator, it then ran a Python script to modify the other applications, a user agent Huntress said differs from the vendor apps it reviewed.

That ambiguity is the governance lesson. If a vendor's normal automation and an attacker using the vendor's credential both look like "the app changed its own roles," then the tenant owner has no way to tell them apart in the moment. The distinction has to be made before the change takes effect, by asking whether that specific role change was routed and approved, not afterward by comparing user agent strings.

What the Logs Cannot Show

The second gap is visibility. Huntress wrote that the access "may have been used for Microsoft Graph directory reads, which leave no audit record." In the tenants it reviewed in detail, it found no mailbox or file manipulation and no user or policy changes. Its honest summary of the state of knowledge was "tokens issued, nothing visible taken."

Microsoft made the same point from the other direction. In its hunting guidance, it cautioned that "the absence of results should not be treated as confirmation that no access occurred." For a defender, that means an empty query is not a clean bill of health. When an identity holds broad read permissions, the logs will show what it changed far more reliably than what it looked at.

Machine Speed on the Other Side

On October 21, Microsoft attributed the activity to Storm-0940, which it describes as a China-state aligned actor active since 2021, and assessed that the group was responsible for the ConnectWise supply chain compromise. Microsoft said the actor used ARTEX, "an open-source penetration testing framework" built as a multi-agent system, and that "Following the unauthorized access, Storm-0940 used ARTEX to attempt to sign-in to more than 2,250 tenants." It also reported "multiple instances where the time from initial targeting to privileged access was less than 3 hours." Microsoft said the majority of the targeting was not successful. The Record, citing Reuters, reported that a Chinese Foreign Ministry spokesperson denied knowledge of the incidents.

Huntress saw the same tempo at ground level. During one stretch on October 15, the app authenticated in tenant after tenant a few seconds apart, in sorted GUID order, which Huntress said points to a script working from a target list. An approval process that runs on a weekly review cycle, or on someone noticing an alert the next morning, is measured against an adversary moving in minutes.

Exposure and recovery windows

Notice came in under two days. Backups took eleven.

Global Administrator grant (Oct 15, 03:44) to vendor advisory (Oct 16, 17:47) ~1.6 days
Backup app access shut off (Oct 17, 05:01) to restoration begins (Oct 28, 16:00) ~11.5 days
Shutoff (Oct 17) to the Nov 4 update, with billing and dashboard still incomplete 18.5+ days

Bars drawn to scale in days on a 19-day axis, from timestamps in the ConnectWise advisory and the Huntress analysis. The exposure bar starts at the Global Administrator grant Huntress observed; earlier role assignments may have been vendor automation. Restoration began on October 28 for re-authorized tenants; a tenant re-authorized later stayed without new backups longer.

The app did not break in. It was already inside, holding the permission to promote itself. What was missing was a person asked to approve the promotion.

Part 3 of 4

The Recovery Gap

Containment Versus Recovery

Huntress's advice to customers was to act immediately: "Disable the application now, even if you see nothing unusual." It also told them plainly what that would cost: "Disabling the app pauses automated backups until it is re-enabled." Their mail, files, Teams messages, and sign-ins kept working, and existing backups held by ConnectWise were not deleted. But no new recovery points were being written.

That is the uncomfortable trade at the center of this incident. The backup service is part of the recovery plan, and its access is broad because backups need to read everything. When that access becomes the threat, the containment step and the recovery capability are the same switch. ConnectWise made the call at the vendor level on October 17, and it was a defensible one. For customers, it meant their backup vendor was effectively offline for Microsoft 365 for about 11 days.

Most business continuity plans assume the backup is there when the primary fails. Fewer plan for the case where the backup provider itself is the incident. If your recovery readiness depends on one vendor's app being connected, then an event that forces you to disconnect it is a recovery event too, and it should already be on the list of scenarios you have walked through.

Re-authorized Is Not Restored

When restoration began, ConnectWise was careful about what a reconnected tenant did and did not mean. "Re-authorization alone does not confirm that backups have resumed," its advisory said. "Verify that new backup snapshots complete successfully." It added that snapshot history would reflect the gap between the last successful backup before the interruption and the initial successful backup after it.

That is good advice, and it generalizes. A green status on a connection shows that an app can authenticate. It does not show that a job ran, that the job captured what you need, or that you can bring it back. Each of those is a separate claim that needs its own evidence, and the strongest evidence is a test restore of real data into a place you can inspect.

Signal after restoration What it shows What it does not show
Tenant re-authorized The new app can sign in to the tenant That any backup has run since the interruption
New snapshot completes A backup job finished for that tenant That it covers the mailboxes, sites, and Teams you depend on
Coverage report reviewed Which workloads are protected and how far back the gap runs That the data can be brought back intact
Test restore performed Real items came back and were checked by a person Whether the tenant itself still carries attacker changes
Tenant role and credential review No unexpected role assignments or new secrets remain What was read through Graph during the exposure window

Why Capable Controls Missed It

None of the affected organizations were short on security tooling, and several of them were watched by a managed detection provider that caught the activity. The issue is where each control sits relative to the change.

Consent grants trust once, then forever

Admin consent for a multi-tenant app is usually a one-time decision. After that, the app's actions inherit the trust, whoever holds its credential.

MFA protects people, not service principals

The app signed in with a client assertion. User-focused controls such as MFA and smart lockout, which locked out two accounts during the password spray, do not sit in that path.

Detection sees the change after it lands

Huntress spotted the role changes and published within about a day and a half. The Global Administrator grant and the new secrets were already in place by then.

Audit logs see writes better than reads

Graph directory reads leave no audit record, so the question of what was viewed may never have a firm answer.

Backup status sees connections, not recoverability

A reconnected app looks healthy. The vendor itself warned that re-authorization alone does not confirm that backups have resumed.

Part 4 of 4

What You Can Do Now

The answer is not to stop using third-party backup or management apps. It is to treat what those apps do with their privileges as changes that need the same scrutiny a human admin's changes would get, and to plan recovery for the day the vendor has to be switched off. These steps apply whether you run a single tenant or manage hundreds for clients.

Seven Practical Steps

1

Inventory vendor apps and what they can do. List every multi-tenant application consented into your tenant, the permissions it holds, and whether any of them can assign directory roles or add credentials to other applications.

2

Route privileged changes made by applications. A service principal adding itself to Global Administrator, or adding a secret to another app, should require a routed change and a named human approver, the same as an admin doing it by hand. Alert on any that bypass that path.

3

Disable, do not delete, during an incident. Huntress and Microsoft both advised disabling affected application IDs rather than deleting them, so that evidence, configuration, and the path back are preserved.

4

Review credentials on every app, not just the vendor app. Disabling an app does not remove secrets it already added elsewhere. Check all applications for new client secrets or certificates, especially long-lived ones, created during the exposure window.

5

Plan for losing the backup vendor. Write down what you would do if your backup provider had to be disconnected for two weeks: an alternate recovery point, an export of critical data, or a second method for the workloads you cannot afford to lose.

6

Validate a restore before trusting it. After any re-authorization, confirm that new snapshots complete for each tenant, measure the gap in recovery points, and perform a test restore of real items that a person checks.

7

Assume reads you cannot see. Where an identity had broad read access during an exposure window, treat directory and data exposure as possible even when the logs are quiet, and decide on notifications and resets on that basis.

Self-Assessment: Ask This About Your Own Tenants

This is not a scorecard of ConnectWise or its customers. It is a checklist for your own organization, whichever vendors you rely on.

? Could a vendor application in your tenant assign itself a directory role today without anyone approving it?
? Would you see, within minutes, a new client secret added to one of your applications?
? If your backup vendor went offline for two weeks, what is your recovery point for mail and files?
? When did someone last restore real data from your Microsoft 365 backup and check it?
? Does your incident plan say who decides to disconnect a vendor, and who owns recovery while it is off?
? Do you know which of your apps can read data in ways your audit logs will not record?

If several of these are hard to answer, the gaps in this incident are already in your environment.

The AuthorityGate take

Two checks were missing here, and neither is exotic. A privileged role or credential change should be a routed change with a named expert in the loop, whether a person, a pipeline, or a vendor's app proposes it. And a restored backup should be validated with a real restore before anyone calls the tenant protected. AuthorityGate Keystone is designed to put both checks in the path and show the evidence behind each decision.

The Bottom Line

A single compromised vendor credential reached Microsoft 365 tenants that had trusted a backup app. Huntress confirmed 162 of them by October 16 and later counted over 300 tenants with successful sign-ins. In at least one tenant, the app made itself Global Administrator in minutes. ConnectWise contained the incident quickly, and the cost of that containment was about 11 days without new Microsoft 365 backups from the service.

The lesson is not that ConnectWise was uniquely exposed. Any vendor app with broad consent carries the same shape of risk, and the response here, from the vendor and from the researchers who caught it, was fast and well documented. The lesson is that trust granted once at consent time keeps working for whoever holds the credential, that privileged changes made by applications deserve a human decision, and that recovery plans need a version where the backup provider is the thing you had to turn off.

This article is part of our AI governance newsletter series. Subscribe to receive complete analyses with timeline tables, governance checklists, and actionable recommendations.

Questions this article answers

What happened in the ConnectWise Cloud Backup incident?

A Microsoft 365 credential (an API secret) belonging to ConnectWise Cloud Backup, formerly SkyKick, was compromised and used to sign in to customer tenants from a hosting provider IP address. Huntress confirmed unauthorized sign-ins in 162 tenants as of 19:00 UTC on October 16, 2026, and later reported almost 1,000 Entra tenants with sign-in activity from that address, over 300 of them successful. ConnectWise shut off Cloud Backup application access on October 17.

Why did backups stop?

Containing the incident meant disabling the vendor applications that run the backups. ConnectWise said shutting off Cloud Backup application access would interrupt backup services for customers until further notice, and Huntress noted that disabling the app pauses automated backups until it is re-enabled. Phased restoration of Microsoft 365 backups through a new application began on October 28, and each tenant had to be re-authorized.

Was customer data stolen?

That is not fully known. ConnectWise said existing backup archives remain encrypted, isolated, and preserved, with no indication of compromise. Huntress found no mailbox or file manipulation in the tenants it reviewed, but noted that Microsoft Graph directory reads leave no audit record, so access to readable data cannot be ruled out. The root cause remains under investigation.

Who was behind the attack?

Microsoft attributed the campaign to Storm-0940, which it describes as a China-state aligned actor, and said the group used ARTEX, an open-source agentic AI penetration testing framework. ConnectWise said that, based on the evidence available to it, it appears a China state-aligned group was responsible. The Record, citing Reuters, reported that a Chinese Foreign Ministry spokesperson denied knowledge of the incidents.

What should Microsoft 365 administrators check now?

Review vendor application role assignments and credentials in Entra, disable rather than delete affected application IDs while preserving evidence, require a named human approver for any privileged role or credential change made by an application, and after re-authorizing a backup vendor, confirm that new snapshots complete and run a test restore for each tenant.

Can a customer rotate the compromised vendor credential?

Huntress described a client assertion held in the vendor tenant. Customers could not rotate that vendor-side credential inside their own Microsoft 365 tenants. Customer-side response instead included disabling affected application access while preserving evidence and reviewing local role and credential changes.

Does re-authorizing the new backup application prove recovery?

No. ConnectWise said re-authorization alone does not confirm that backups have resumed. The article recommends confirming new snapshots and completing a test restore for each tenant before treating backup protection as restored.

Why are vendor privilege changes an AI governance issue?

The article describes a vendor app assigning directory roles without a customer-side human decision, alongside Microsoft reporting ARTEX use in the campaign. Its governance recommendation is to treat role and credential changes as reviewable changes, with named approval and evidence before expanded authority is trusted.

Sources

Tenant counts come from different observers and are not directly comparable. The 162 and over-300 figures are Huntress's, drawn from tenants where it could see sign-in logs; the 2,250-plus figure is Microsoft's count of attempted sign-ins. No source has published a total number of compromised tenants. Huntress said the role assignments in roughly 50 tenants carried a user agent consistent with ConnectWise's own automation but that it had not confirmed this with ConnectWise. The 2,900 failures per minute figure describes a password-spray campaign that Microsoft linked to the same network range; it is not a count of successful sign-ins. Whether data was read through Microsoft Graph cannot be ruled out because those reads leave no audit record. The root cause, including how the credential was obtained, remains under investigation. Attribution is Microsoft's assessment; a Chinese Foreign Ministry spokesperson denied knowledge of the incidents, as reported by The Record citing Reuters. ConnectWise said on October 21 that it had begun restoring services in controlled phases, but partner-facing Microsoft 365 backups returned through the new application on October 28. The roughly 11-day backup gap runs from the October 17 shutoff to that October 28 restart; individual tenants may have resumed later depending on when they were re-authorized.