A backup product is supposed to be the boring thing in your stack. It copies mail and files somewhere safe, you never think about it, and the only time it earns its licence fee is the day someone fat-fingers a mailbox purge. That’s the deal.
The deal has a catch nobody reads at onboarding. To back up a Microsoft 365 tenant, the product needs an application identity, a service principal with standing, app-only permissions to read every mailbox and every SharePoint site, no user present, no MFA prompt, no conditional access in the way. Multiply that across the hundreds or thousands of tenants an MSP manages through one console and you’ve built a master key ring. ConnectWise just discovered what happens when someone copies the ring.
The company shut down customer access to its Cloud Backup service after finding that attackers had compromised application identities and used them to reach into customer Microsoft 365 tenants. Not a hypothetical. Not a proof-of-concept in a lab. Live access to customer data, which is why the response was to turn the whole thing off rather than issue a patch note and a shrug.
Let me walk the clock, because the sequence tells you more than any single alert would.
The part where it was called SkyKick
Cloud Backup didn’t start life at ConnectWise. It was SkyKick Cloud Backup, a well-regarded M365 backup tool that MSPs bolted onto their stacks for years. The design predates the acquisition, and the design is the story: to protect a tenant, an admin consents the backup application into it. That consent grants app-only Graph and Exchange permissions, the kind that let a daemon read mail at 3am without anyone logged in.
This is not a SkyKick flaw or a ConnectWise flaw. It is how every SaaS-to-M365 backup product on the market works. The architecture is fine right up until the vendor’s half of the trust relationship gets breached, at which point the architecture is a loaded gun pointed at every tenant that said yes.
The part where ConnectWise bought it and rebranded it
The product became ConnectWise Cloud Backup. New name, same consent grants already sitting in your tenants. This matters for one unglamorous reason: the service principal in your Azure AD directory may still carry a display name or app ID from the SkyKick era, or a generic backup name, or something that doesn’t obviously say “ConnectWise” at all. When you go looking, and you should, today, don’t grep for one string and call it clean.
Every rebrand leaves this kind of sediment. The thing you consented to in 2021 under one vendor’s name is the thing with god-mode on your mail in 2026, and the sticker on the box changed twice in between.
The part where someone compromised the application identities
This is the beat that matters. Attackers didn’t phish your admins or brute-force your users. They compromised the application identities, the service principals tied to the backup integration, and used them to access customer Microsoft 365 tenants directly.
Think about what that bypasses. App-only authentication carries no user, so there’s no password to steal and no MFA to defeat. Conditional access policies that key off user risk, device compliance, or named locations frequently don’t apply to app-only flows at all unless you’ve explicitly built workload identity policies, and almost nobody has. The access shows up in your logs as the legitimate backup app doing legitimate backup things. That’s the ugly part. A compromised backup identity reading your mailboxes looks almost exactly like a backup identity backing up your mailboxes.
Which M365 data was in reach depends on the permissions the integration was granted, but a backup product’s shopping list is predictable and broad: full mailbox read across Exchange Online, SharePoint and OneDrive content, and the directory metadata to enumerate who’s who. That’s email, documents, and user data, at the tenant level, for every tenant that onboarded. ConnectWise has not, per the reporting, said every customer was accessed, and you shouldn’t assume you were. You should assume you could have been and prove otherwise from your own logs.
The part where ConnectWise turned it off
ConnectWise disabled customer access to Cloud Backup entirely. Blunt, correct, and expensive, pulling a backup product offline means customers lose their backups and their restore path for the duration. The company chose that over leaving the integration live, which tells you how much they trusted the compromised identities to stay contained. When a vendor eats that much operational pain voluntarily, believe the severity.
But their containment protects their platform. It does not touch the consent grant and credentials sitting in your tenant. A disabled service doesn’t revoke the service principal’s standing access, doesn’t delete any client secrets or certificates attached to it, and doesn’t invalidate tokens already minted. That cleanup is on you, and it’s the whole job this week.
What to actually check, right now
Find the service principal first. Read-only, no changes, look before you touch.
Once you have the object ID, pull what it can actually do and what credentials are hanging off it. App role assignments are the app-only Graph permissions; the OAuth2 grants are delegated ones. Credentials are where a quietly-added attacker secret would hide.
Any client secret or certificate you don’t recognise, especially one added recently, is your signal. Note the KeyId and the start date and compare against your own change records.
Now the behaviour. App-only sign-ins live in a separate log from interactive ones, and this is the query most people miss:
Then go to the content layer via the unified audit log, because mailbox and file access leaves tracks Graph sign-ins don’t. Look for the backup app doing things outside its backup window, from IPs that aren’t the vendor’s:
Baseline the source IPs against ConnectWise’s published ranges for the service. Backup traffic from the vendor’s infrastructure is expected. Mailbox reads from anywhere else, under the app identity, are not.
What to revoke, and the thing not to do
Prioritise in this order. First, rotate or remove any credentials on the application object you can’t account for, that kills the attacker’s ability to mint new tokens. Second, if you’ve confirmed the service is down on ConnectWise’s side and you have no live restore need, revoke the app role assignments or disable the service principal’s sign-in (AccountEnabled = $false) to sever standing access entirely. Third, revoke existing refresh tokens for the principal so already-issued access can’t be replayed.
Here’s what not to do: do not write a loop that bulk-nukes every app registration matching “backup” across every managed tenant at 2am. You will break legitimate integrations you still depend on and turn one incident into fifty. Work tenant by tenant, confirm each service principal is the one in scope, and keep the report output from the scripts above as your record of what you changed and why. One tenant, verified, then the next.
Urgency: immediate. This isn’t a patch-next-cycle item. App-only access is live until someone revokes it, the credentials are valid until someone rotates them, and the access pattern hides inside expected backup behaviour. Every day the consent grant sits there unreviewed is a day you’re trusting that the attackers who got into the vendor can’t get back in.
The next dated beat is ConnectWise’s restoration of the service, whenever that lands, with whatever new app IDs and credentials it ships. When it does, treat the re-consent as a fresh grant to scrutinise, not a formality to click through. The permission scope a backup product needs hasn’t shrunk, and the master key ring gets rebuilt the moment you say yes again. Read it this time.
