top of page

Why the Most Privileged Identity in Your Microsoft 365 Tenant May Not Be Human

Kostas Tsiolas
Oct 19
9 min read

What the ConnectWise Cloud Backup incident teaches us about third-party application risk—and how to find unmanaged app identities before an attacker does.


Facts as of 18 October 2026. The investigation is ongoing. Check the ConnectWise advisory for the latest status before acting on any detail below.

Executive summary

Between 13 and 15 October 2026, an attacker used a stolen credential belonging to ConnectWise Cloud Backup (formerly SkyKick) to access at least 173 Microsoft 365 tenants. There was no phishing email, guessed password or MFA prompt. The attacker authenticated as an application that customers had already authorised to read mail and files.

This is not simply a story about one supplier. ConnectWise is the second Microsoft 365 backup provider in under two years to report abuse of application credentials. Leadership teams should ask: which third-party applications can read our tenant, who owns each one, and how quickly could we disable one and prove what it accessed?

Three actions to take this week

  • Inventory and assign owners. List every third-party app holding tenant-wide permissions and give each one an internal owner.

  • Remove dead consents. Remove consents for tools you no longer use.

  • Watch the apps. Make sure someone reviews service-principal sign-ins, and enable Microsoft Graph activity logs.

Below, we explain the incident, why standard controls offered limited protection, and the practical checks your team can run today.


What happened

ConnectWise Cloud Backup is a Microsoft 365 backup service delivered largely through managed service providers (MSPs). To perform backup and recovery, it receives broad application permissions across customer tenants, potentially covering mailboxes, OneDrive, SharePoint, Teams and directory data. Microsoft said the permissions likely abused included Directory.ReadWrite.All.

The credential proving the application’s identity sits in the vendor’s tenant. The same credential can operate across every customer tenant that has consented to the multi-tenant application. That is normal platform design—and it is the trust path the attacker exploited.

Incident timeline (UTC; compiled from Huntress, Todyl, Microsoft and ConnectWise)

Date

Event

13–14 Oct

The attacker's IP address password-sprays four law-firm tenants. No successful user sign-ins observed.

14 Oct

ConnectWise identifies the issue internally.

15 Oct, from 01:58

In about 50 monitored tenants, ConnectWise Cloud Management apps assign themselves directory roles, mostly Exchange Administrator. Huntress notes this may be vendor automation; it is unconfirmed.

15 Oct, 03:44–03:49

In one tenant, the Cloud Management – MS Teams app adds itself to Global Administrator, then adds new client secrets to two other trusted applications: an MSP management app and a security tool's app.

15 Oct, 06:13–08:34

The Cloud Backup app requests Microsoft Graph tokens from a hosting-provider IP across customer tenants. It works through them in sorted tenant-ID order, seconds apart, which points to a script and a target list.

15 Oct, 09:25–14:20

A second, targeted wave from the same IP hits 16 tenants in the policy, national security, trade and critical-minerals sectors. It is unsuccessful; most of those tenants don't use ConnectWise products.

16 Oct

ConnectWise publishes its advisory. Huntress confirms 162 affected tenants. Todyl publishes its investigation.

17 Oct, 07:00

ConnectWise disables seven application identities and revokes certificates. The Huntress count reaches 173. Microsoft confirms it is tracking an active supply chain attack.

18 Oct

ConnectWise says the shared credentials are no longer valid. Backups and restores remain paused. The root cause is still under analysis.

Three important caveats

  1. 173 is a lower bound. It is what one security vendor could see. Huntress also saw almost 1,000 tenants with sign-in activity from the attacker's IP, over 300 of them successful.

  2. No one has confirmed data theft. That is partly because, as the next section shows, the logs that would prove it often don't exist.

  3. The two phases may not be linked. Huntress has not tied the early role changes and the later token sweep to the same operator. Treat both as things to check in your tenant, not as a settled story.

 

Why familiar security controls did not help

Go through the controls most organisations rely on and ask whether each one covered this identity.

MFA. App-only sign-ins involve no user and no device, so there is nothing to prompt. Strong human authentication, even phishing-resistant passkeys, doesn't touch this path.

Conditional Access. Microsoft offers Conditional Access for workload identities, but documents that it applies only to single-tenant service principals registered in your own tenant, and it needs Workload Identities Premium licensing. Third-party SaaS and multi-tenant apps are out of scope. You cannot put a location block on your backup vendor's app, however much you want to.

Sign-in monitoring. Application sign-ins appear in a separate tab in Entra (Service principal sign-ins), not in the default user sign-ins view. If your SOC or MSP watches only user sign-ins, this activity was invisible.

Session revocation. Revoking user sessions does nothing here, because no user session exists. Disabling the app in your tenant stops new tokens being issued, but tokens already issued stay valid until they expire. Todyl puts that window at up to 90 minutes.

Audit trail. Microsoft's directory audit log records role assignments and credential changes. For app-only actions, though, it doesn't record a source IP, and it doesn't record what an application read through Microsoft Graph. Where legitimate backups were running, Todyl could not tell the vendor's reads apart from the attacker's. Both used the same identity, and the Graph activity logs that carry per-request source IPs weren't being collected.

The outcome: one of the identities with the broadest access in the tenant was also among the least protected by authentication, policy, monitoring and logging.

 

The deeper issue: privileged identities without owners

Remove the vendor name and the underlying issue is a governance gap.

When an organisation onboards a SaaS tool, an administrator, often at the MSP, clicks Accept on a consent screen. From that moment, a service principal sits in the tenant holding whatever the vendor asked for. It has no manager. It isn't part of joiner-mover-leaver processes or quarterly access reviews. And when the tool is replaced, the consent usually stays behind.

ConnectWise has asked partners to check environments with dormant SkyKick connections. These are tenants that may have stopped using the product years ago and were still exposed.

In almost every Microsoft 365 tenant I review, the attention goes to human identities: MFA coverage, admin roles, guest access. Third-party application permissions, leftover consents from old tools and service-principal sign-ins are almost never checked. This isn't negligence. No process asks anyone to look.

There is also a limit to what any vendor can see, and no contract changes it. ConnectWise stated that its own monitoring cannot establish the status of customer environments, which by design it cannot observe directly. The vendor protects the credential. Only you can see what that identity did inside your tenant. If you aren't looking, nobody is.

And this isn't a one-off. In 2025, CISA warned that threat actors may have accessed client secrets for Commvault's Metallic Microsoft 365 backup service, which gave them access to customer tenants. CISA assessed that this could be part of a wider campaign against SaaS applications with default configurations and elevated permissions. That makes two backup vendors with the same pattern in under two years. Backup and management tools are attractive targets precisely because they need the broadest access.

 

If you used ConnectWise or SkyKick, act now

If any ConnectWise or SkyKick application was ever consented in your tenant, or in tenants you manage, work through these steps. Preserve logs before you change anything.

1. Preserve evidence.

  • Export service-principal sign-in logs and directory audit logs from 1 October onward.

  • Entra ID Free keeps sign-in logs for only seven days.

2. Contain.

  • In the Entra admin center, go to Enterprise applications and set the application type filter to All applications.

  • Search for "SkyKick" and "ConnectWise".

  • For each app, open Properties and set Enabled for users to sign-in? to No.

  • Disable; don't delete. Deleting removes the consent and makes recovery harder.

The seven application IDs ConnectWise disabled:

Application

Application ID

ConnectWise Cloud Backup (formerly SkyKick Backup)

506e08f9-6e9a-4da4-942b-a099a03644ae

ConnectWise Cloud Management

2399d300-6fca-4e6a-a686-399fefe75b83

Cloud Management – MS Teams

24205314-2b89-4e07-a2a4-579e898a5640

Cloud Management – Exchange Online

b78f6a77-d4e5-46fb-87e5-d1648bd02519

Cloud Management – MS Graph

0de14c77-b9b0-4137-9360-34878a929827

Cloud Management – Partner Center

412dcf96-176d-4b3a-a1b0-4981ad66971e

Cloud Management – SharePoint Online

b2b55f5b-7758-4719-8d8f-88ae172e6bdb

In older tenants, the Cloud Management apps may show as "SkyKick Cloud Manager".

3. Record exposure. Before changing permissions, export each app's granted permissions. That defines what could have been reached.

4. Hunt. Check for:

  • Service-principal sign-ins for these app IDs from anywhere other than Microsoft's internal IPv6 space, and specifically from 144.172.64[.]128.

  • Directory audit events where any ConnectWise or SkyKick app added itself to a role, added credentials, added owners or granted consent.

  • Credentials (secrets and certificates) added since 13 October on all applications and service principals, not only the ConnectWise ones. In the observed case, the planted secrets were on other trusted apps.

  • Mailbox forwarding, inbox and transport rules, and delegated mailbox access.

The queries are in the appendix.

5. Remove persistence. Remove any credential, role assignment, owner or consent you can't attribute to a known change. Disabling the vendor app doesn't remove credentials planted on other applications, and ConnectWise notes that disabling connections doesn't reverse changes already made.

6. Don't re-enable early. Keep the apps disabled until ConnectWise publishes reconnection steps. When you re-enable, check whether the permissions requested are still what you need.

7. Manage the backup gap. Scheduled backups aren't running. Establish your last good recovery point and decide whether any business-critical data needs interim protection.

8. If you are an MSP, check your own tenant first, then your management and security tools. Those apps are usually consented into all of your customers' tenants.

 

Self-assessment: does this gap exist in your tenant?

You do not need to be a ConnectWise customer for this risk to apply. Answer these questions honestly: every “no” or “don’t know” is a finding.

  1. Do we have a current list of every third-party application holding application permissions (not just delegated) in our tenant?

  2. For each one, do we know which permissions it holds, especially tenant-wide read or write access to mail, files, Teams or the directory?

  3. Does each one have a named internal owner who can explain why it exists?

  4. Have we removed consents for tools we no longer use, including tools installed by a previous IT provider?

  5. Does any third-party service principal hold a directory role, such as Exchange Administrator or Global Administrator? Can someone explain why?

  6. Is someone (internal team, SOC or MSP) actually reviewing service-principal sign-ins, not just user sign-ins?

  7. Are Microsoft Graph activity logs enabled and kept long enough to answer "what did this app read?"

  8. Would we notice if a new secret or certificate were added to an existing application?

  9. Could we disable a compromised app within minutes? Has anyone ever practised it?

  10. Do our vendor contracts say how quickly the vendor must tell us about a credential compromise, and what access the vendor holds?

Finding an application nobody can explain is not a failure. It is the purpose of the exercise—and most organisations will find more than one.


What good application identity governance looks like

A living inventory. Record every third-party app with application permissions, with its publisher, permissions, business purpose and owner. Review the list every quarter, and whenever a vendor contract starts or ends.

Real ownership. Each app has an internal business owner and a technical owner. If nobody will own it, it doesn't stay.

Least privilege at consent time. Ask vendors to justify tenant-wide permissions. Where a product supports narrower access (specific mailboxes or sites rather than all of them), use it. Prefer vendors that document exactly which permissions they request and why.

Control over who can consent and change apps. Restrict user consent and route requests through the admin consent workflow. Limit who holds Application Administrator and Cloud Application Administrator, because those roles can add credentials to apps.

Credential hygiene for apps you own. For applications registered in your own tenant:

  • Prefer certificates or managed identities over client secrets.

  • Limit credential lifetimes with application management policies (check licensing).

  • Apply Conditional Access for workload identities where licensing allows.

  • If you publish multi-tenant apps yourself, enable app instance property lock.

Monitoring built for non-human identities. Service-principal sign-ins and app credential changes feed your SIEM, with alerts on new source IPs, new credentials and roles an app assigns to itself. Microsoft Graph activity logs are enabled. They require Entra ID P1 or P2 and are not retroactive, so the time to enable them is before you need them.

A tested kill switch. A runbook names who can disable which app, how quickly, and what business impact follows (for example, backups pausing). Rehearse it once as a tabletop exercise.

Contracts that match the access. For any vendor whose app can read your whole tenant, require:

  • Breach-notification deadlines that cover credential compromise.

  • Disclosure of the permissions the app requests.

  • An offboarding clause that includes removing the consent.


Three questions for your next leadership meeting

This incident will leave the news cycle. The pattern will not. Every Microsoft 365 tenant contains applications with extensive access, fewer safeguards than human users and, too often, no accountable owner.

Ask these three questions:

  1. Which third-party apps can read all our mail and files?

  2. Who owns each one?

  3. Could we switch one off in minutes and prove what it touched?

If your self-assessment uncovers an application nobody can explain, talk to Nimbus Cyber. That is often where the most valuable identity-security work begins.

Nimbus Cyber helps organisations across Europe govern human, non-human and AI agent identities. Explore our identity security services.


Sources


Comments


bottom of page