All postsCybersecurity

Microsoft breach 2026: why MFA did not stop Storm-0940

APJ-RAISE team · · 8 min read

The biggest "Microsoft breach" of late 2026 was not a break-in at Microsoft. A China-aligned group that Microsoft tracks as Storm-0940 stole the credentials of ConnectWise backup apps and used them to sign in to customer Microsoft 365 tenants, where MFA never had a chance to stop it.

Rows of black server racks in a data center
Photo: Brett Sayles, Pexels

That last part puzzled many readers. Most of us were told for years that turning on MFA closes the door. So this post starts from one research question: why did MFA not matter here? The answer says a lot about where cloud attacks are going.

Was there a Microsoft breach in 2026?

As of 18 November 2026, nobody has reported a breach of Microsoft's own corporate network in October or November. In its 21 October report, Microsoft said the Storm-0940 activity did not come from a flaw or compromise in Microsoft systems. The way in was a third-party product, ConnectWise Cloud Backup, which many companies knew by its old name, SkyKick.

Customers connect a backup service like this to their tenant once and then forget about it. A tenant is simply one organisation's own space inside Microsoft 365. The attackers did not need to fool any employee of those organisations. They only needed the keys the backup app already held.

ConnectWise breach: what happened to SkyKick customers

According to Microsoft, Storm-0940 took over the identities of several ConnectWise cloud apps and then tried to sign in as those apps to more than 2,250 tenants. The first malicious app activity Microsoft looked at dates to 15 October 2026. In some victims the attackers reached privileged access less than 3 hours after first contact.

Once inside, they did things a normal backup app has no reason to do. They mapped the directory through Microsoft Graph. They added fresh credentials to OAuth apps that already existed in the tenant, which works as a hidden back door. One ConnectWise app was given the Global Admin role. Azure Key Vault secrets were read in bulk.

Microsoft also says the group used ARTEX, an open-source AI pentest framework, to run much of this automatically. We wrote a separate explainer on how the ARTEX agent framework works.

Before the app logins, there was a plain password spray from 10 to 14 October. It hit 327 accounts in 8 organisations and produced over 151,000 failed logons, peaking near 2,900 failures a minute on 11 October. Microsoft saw no successful sign-in from it. Our own arithmetic: that is roughly 460 wrong guesses per account. MFA and lockout rules did their job against the spray. The trouble came from a different door.

Why MFA is not working against app sign-ins

MFA asks a human to prove they are present, usually with a phone tap or a passkey. An app that runs on a server has no phone and no fingers. So Microsoft 365 lets apps prove who they are in another way, with a client secret or a certificate. Steal that proof and no second factor stands in the way.

This text diagram shows the two paths side by side.

  • User sign-in: person types password, then Entra ID asks for MFA, then phone or passkey answers, then the user gets a token.
  • App sign-in: server sends app ID plus secret or certificate, then Entra ID checks the secret, then the app gets a token. No MFA step exists anywhere on this path.

Now picture one vendor app used by thousands of customers. Each customer clicked "consent" once, which created a service principal (the app's local identity) in their tenant with the permissions the app asked for. Backup apps ask for a lot, because they have to copy mail, files and chats. Steal the vendor's secret and you can knock on every one of those tenants with the same key.

A quick note on sourcing. Microsoft's 21 October report describes the app identity abuse but does not use the word MFA. The MFA point is our explanation of how app-only sign-in works, and BleepingComputer made the same point in its coverage.

User credentials vs app credentials

We could not find a simple side-by-side comparison in the coverage we read, so we made one:

  • Who uses it: a user account belongs to a person; an app credential belongs to software.
  • Proof: a person shows a password plus MFA; an app shows a secret or a certificate only.
  • Expiry: user sessions are checked often; app secrets can stay valid for months or years.
  • Where logs go: user sign-ins appear in the usual sign-in log; app logins sit in a separate service principal log many teams never open.
  • Reach: one user reaches one mailbox; one multi-tenant app can reach every customer that consented.
  • Who reviews it: HR removes people who leave; few teams have anyone who removes old apps.

Read that list again and the attack makes sense. App identities are strong, quiet and rarely checked. Calling this an MFA bypass is a little unfair to MFA. The attackers simply used a door that never had MFA on it.

How to check your own tenant for app-identity risk

If you help run a Microsoft 365 tenant for a college lab or a small company, you can do a short review in the Entra admin center without any special tools. We describe the idea, not exact clicks, because menus change.

  1. 1Open the list of enterprise apps and sort by permissions. Note any app that can read or write all mailboxes or all files.
  2. 2For each of those apps, ask: who in our team asked for it, and do we still use it? If no one knows, mark it.
  3. 3Look at the credentials on your own app registrations. A secret added recently that nobody remembers adding is the exact pattern Storm-0940 left behind.
  4. 4Open service principal sign-ins, which sit apart from user sign-ins, and look for apps signing in from unusual places or at unusual hours.
  5. 5Check who holds Global Admin. Any app on that list needs a very good reason.

If something looks wrong, save the logs before you delete anything. Then remove the extra credentials and rotate any secrets the app could see, including Key Vault secrets. Our guide to identity for AI agents covers the same least-privilege idea for agent software.

What most coverage got wrong

Two numbers got stretched as the story travelled.

First, 2,250 is a count of tenants that received sign-in attempts. Some headlines read as if all 2,250 were breached. Microsoft has not said how many customers actually lost data. The only firm count we found is from Huntress, which reported at least 162 compromised tenants among its own customers.

Second, the AI angle. A separate Microsoft report from 5 November describes an AI agent going from a leaked GitHub secret to Key Vault data in five minutes. Several write-ups folded that case into the Storm-0940 story, but that report does not name Storm-0940. Treat them as two cases that show the same trend.

What else was called a Microsoft breach in November 2026?

Three other stories ran under the same headline. None of them involved a break-in at Microsoft, but each one needs its own fix.

Outlook zero-day. CVE-2026-124194, patched on 10 November 2026, is an Outlook spoofing flaw that attackers used before the fix. A crafted message file makes Outlook contact an attacker server with no external-content warning, which leaks the user's NTLMv2 hash. It is rated Moderate, CVSS 6.5, and CISA put it on its exploited list the same day. Microsoft credits "nono" and "Benoît Sevens using Claude with Anthropic" for finding it. The same release fixed CVE-2026-116928, a Critical (CVSS 9.8) Windows Search bug that fires when Exchange indexes an attachment, not exploited at release. Counts for the month differ: SecurityWeek says 1,094 Microsoft CVEs, BleepingComputer says 1,235, so "more than 1,000" is the safe phrase.

The outage. Outlook, Teams, SharePoint, Copilot and Defender XDR went down from about 1:26 PM US Eastern on 12 November, close to midnight in India. Microsoft called it a technical issue not related to cyber or AI, traced to an earlier network capacity upgrade. Recovery came around midday Eastern on 13 November.

The fake outage emails. From 13 November, Microsoft saw a high-volume campaign copying its service alerts, which it judged likely credential phishing. The subject line reads "Outage" plus a random token, sender names look like "MSF Outage Notification", and a button such as "RESTORE MESSAGES" warns that mail will be deleted in 24 hours. Real Microsoft alerts live in the admin center. When in doubt, report the email to your IT team and do not click.

For attacks aimed at AI apps themselves, our OWASP Top 10 for LLM applications post is a good next read, and India's view of AI-driven attacks is in our note on the CERT-In CRAFTY FALCON advisory. Students who want to study app-identity abuse as a project can start with our research foundations course.

Questions people ask

Can MFA be bypassed?
Yes, in some cases. In the Storm-0940 attack, nobody bypassed MFA directly; the attackers signed in as an app with a stolen secret, and app sign-ins never ask for MFA.
What is service principal authentication?
It is how an app proves its identity to Microsoft Entra ID, using a client secret or a certificate instead of a password and MFA. Each tenant that consents to an app gets its own service principal for it.
Is ConnectWise safe after the Cloud Backup breach?
ConnectWise took several cloud products offline in October 2026, according to news reports. If your organisation used Cloud Backup or SkyKick, review that app's credentials and permissions in your tenant before trusting it again.

Sources

  1. 1.Microsoft Threat Intelligence. AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise. Microsoft Security Blog, 21 October 2026.
  2. 2.Microsoft Threat Intelligence. Threat actors use agentic AI tooling to find and abuse exposed secrets. Microsoft Security Blog, 5 November 2026.
  3. 3.Microsoft Security Response Center. CVE-2026-124194: Microsoft Outlook Spoofing Vulnerability. 10 November 2026.
  4. 4.Microsoft Security Response Center. CVE-2026-116928: Windows Search Remote Code Execution Vulnerability. 10 November 2026.
  5. 5.CISA. Known Exploited Vulnerabilities Catalog.
  6. 6.Microsoft Threat Intelligence. High volume phishing campaign uses Microsoft 365 outage lures to target multiple industry sectors. Microsoft Security Blog, 16 November 2026.
  7. 7.Lawrence Abrams. Massive Microsoft 365 outage disrupts Outlook, Teams, Copilot worldwide. BleepingComputer, 12 November 2026.
  8. 8.Eduard Kovacs. Exploited Outlook Zero-Day Among 1,094 Flaws Patched by Microsoft. SecurityWeek, 10 November 2026.
  9. 9.Ionut Ilascu. Hackers breach Microsoft 365 tenants via ConnectWise supply-chain attack. BleepingComputer, 17 October 2026.
  10. 10.Carly Page. Microsoft says ConnectWise breach let Chinese spies loose on 2,250 cloud tenants. The Register, 22 October 2026.
All posts