109 days
the flaw sat in the Connect Health API before Microsoft detected it
10 days
from detection to the server side fix
19 Nov
Microsoft deployed the fix; no customer patch is needed
5
Microsoft says keep Global Administrators to fewer than five
The stolen data is a target list, not a secret
Nothing in this breach would open a single account. That is the comfort in Microsoft's notice, and it is the wrong one. What was read from the Entra Connect Health API is the list an identity attack is built from: the addresses of the people who administer the tenant, the tenant's name and domain, and whether it trusts an on premises Active Directory.
Every identity engagement we run starts by building that list by hand. Who holds the privileged roles. What their sign in names look like. Whether the tenant is cloud only or hybrid, because a hybrid tenant has a second door, the directory on the ground, and the sync that carries its accounts upward. That work takes time, it is noisy, and it is often incomplete. An attacker who read this API got a cleaner version of it, for some set of tenants, straight from the source.
So the question for an Entra customer this week is not whether data was lost. Microsoft says the read was limited and that no customer remediation is required, and on the narrow question of the bug that is right: the fix is on Microsoft's side and it is in. The question is what the people on that list can be talked into, and whether the accounts they hold will survive a well researched phishing attempt or a convincing call to the help desk.
If the honest answer is that some Global Administrators still sign in with a password and a push notification, the breach did not create that weakness. It just removed the guesswork for whoever comes next.
What Microsoft says the Connect Health flaw exposed
Microsoft Entra Connect Health is the monitoring service for hybrid identity. In Microsoft's description it "provides monitoring for your on-premises identity infrastructure", and it works by installing an agent on each on premises identity server: the Entra Connect Sync server, AD FS and the domain controllers. The portal brings that health data together in one place. It needs Entra ID P1 licensing, so the tenants that use it are, almost by definition, organizations with a real on premises directory and a paid identity tier.
BleepingComputer reported on 21 Nov 2026 that Microsoft is notifying customers about a vulnerability in the Connect Health API. Per the notification it quotes, Microsoft detected the flaw on 9 Nov 2026, the flaw had been introduced on 23 Jul, and an investigation found it had been exploited. Microsoft's own wording to affected customers is that "unauthorized read-only access to tenant-level metadata associated with your organization had occurred", and that it "did not identify any unauthorized configuration changes". The report adds that the flaw could potentially have allowed changes to Connect Health monitoring configurations, which is why reviewing that configuration is part of the advice.
The fields the notification lists as exposed:
-
Service //
Microsoft Entra Connect Health API, run by Microsoft -
Live //
introduced 23 Jul 2026, detected 9 Nov 2026 -
Fixed //
server side, 19 Nov 2026; no customer patch -
CVE //
none published -
Exposed //
admin and technical contact email addresses -
Also exposed //
tenant ID, display name, initial domain, country and region -
Also exposed //
license tier, sync enabled and last sync time, Connect Health onboarding status -
Not exposed //
passwords, hashes, tokens, valid keys, AD topology, Azure subscriptions -
Exploited //
yes, per Microsoft's notice
One more item is on the list: an expired certificate that had been used to register Connect Health agents with a Microsoft internal service decommissioned in late 2024. Microsoft does not present it as usable, and we have no reason to think it is. It is worth knowing about because it tells you what the API had in reach.
Microsoft deployed a server side fix on 19 Nov, ten days after detection, and the investigation closed on 20 Nov, when notifications began. Microsoft told BleepingComputer: "We identified and addressed a security issue affecting Microsoft Entra Connect Health." It has not said who exploited the flaw, how many tenants were affected, or when the activity began. Administrators started posting the notice on Reddit, more than a dozen of them by the time of the report, and some managed service providers said they had received it for several customer tenants at once.
The notice also points to a set of read only PowerShell reference scripts that Microsoft publishes at AzureAD/EntraIDGovernance on GitHub, which "inspect the information returned for your tenant by the Microsoft Entra Connect Health APIs". There is a catch the report itself makes plain: the scripts show the current values in your tenant, not what was read during the breach. They tell you the shape of what an attacker could have seen. They do not tell you what an attacker did see.
How tenant metadata turns into an admin takeover
Read the field list the way an operator does and most of it sorts into three questions. Who should we target? What will they believe? Where is the second way in?
The administrator and technical contact addresses answer the first. They are the accounts that matter, and in many tenants the address and the sign in name are the same string. The display name, initial domain, country and region answer the second: they are the details that make a pretext sound like it came from inside, or from Microsoft. The license tier and synchronization status answer the third. A tenant with sync enabled and a recent last sync timestamp is running a live bridge from an on premises directory, and that tells an attacker the on premises side is worth attention, because accounts that start there end up in the cloud.
Tenant IDs and initial domains on their own were never secrets. Microsoft serves each tenant's OpenID Connect configuration document at a publicly accessible endpoint, addressed by the tenant's domain name or its tenant ID. The new part is the pairing: an identifier that is public, joined to the names of the people who hold the keys, joined to a fact about the architecture you would otherwise have to work out.
Figure 01
From a metadata read to the tenant
- 01
Read //
The API returns tenant names, domains, sync status and the admin addresses.
- 02
Aim //
Admins become a named list, with a pretext built from the tenant's own details.
- 03
Ask //
A proxied login page or a help desk call targets one admin account.
- 04
Hold //
A captured session or reset method gives the attacker an admin role.
- 05
Spread //
Admin rights reach mail, apps and, in a hybrid tenant, the sync path.
Result //
Step one leaked no secret. Steps three and four decide the outcome.
Nothing in that chain needs a second Microsoft bug. Step three is the standard repertoire we have written about before: proxying a login so the user passes MFA and the attacker keeps the session, device code phishing, and the help desk reset. Each of them works far better against a named administrator with a believable pretext than against a guessed address. Step four is where a stolen session outlives a password reset, which is why the response to a targeted admin has to include revoking sessions, not only changing a password.
The controls that break the chain sit at step three and four, and they are not new. Microsoft's built in phishing resistant authentication strength allows Windows Hello for Business or a platform credential, FIDO2 security keys, and Entra certificate based authentication. Those methods bind the sign in to the real sign in surface, so a proxied page cannot relay them. An admin who can only sign in that way is not much use to an attacker holding their email address.
What the notice does not say
Microsoft has been specific about the data and quiet about almost everything else. Four gaps matter for how a customer should respond.
How many tenants, and which. Microsoft has not said. Notices are arriving per tenant, which is the right mechanism, but the absence of a count means a customer who did not receive one cannot tell whether that means "not affected" or "not yet told". Our working assumption for any tenant that uses Connect Health is that its admin addresses should be treated as known.
When the activity started. The flaw was live from 23 Jul to 9 Nov, which is 109 days. Microsoft has not said when the first unauthorized read happened. If the reads began early in that window, any targeting built on them may already have happened, and the place to look is sign in history going back to the summer, not just the last week.
Who did it. No attribution, no indicators. That is common at this stage, and we would rather Microsoft say nothing than guess. The practical consequence is that there is no actor profile to hunt for. The hunt has to be built on behavior: what someone holding your admin list would do next.
Why there is no CVE. This was a flaw in a service Microsoft runs, fixed on Microsoft's side, so there is nothing for a customer to install and no CVE has been published. That is consistent with how cloud service flaws are often handled. It also means the incident will not show up in a vulnerability scanner, a patch report or a KEV feed. The only record many organizations will have is a Service Health message that one administrator read. Make sure it reached the security team.
How to tell if your tenant is being targeted
Start with what Microsoft gives you. If your tenant received the notice, keep it, and record the date it arrived. Run the reference scripts to see what the Connect Health APIs return for your tenant today, so you know which admin and contact addresses were in reach. Then review the Connect Health settings themselves: who holds the Owner, Contributor and Reader roles, and where alert mail goes. Connect Health email notifications are on by default, and the recipients can be set to all Global Administrators plus any custom addresses, so the configuration is also a list of who gets told when hybrid identity breaks.
Next, look for the behavior that would follow from someone holding your admin list. Two patterns are worth a query: a rise in failed sign ins against admin accounts from addresses you have not seen before, and successful admin sign ins that satisfied only single factor authentication. The first is targeting. The second is the gap the targeting is looking for.
Figure 02
Admin accounts under pressure
KQL
// Replace with your privileged accounts, or load them from a watchlist
let Admins = dynamic(["admin1@contoso.com", "admin2@contoso.com"]);
let Lookback = 30d;
// 1. Failed sign ins against admin accounts, grouped by source
SigninLogs
| where TimeGenerated > ago(Lookback)
| where UserPrincipalName in~ (Admins)
| where ResultType != "0"
| summarize Failures = count(), Apps = make_set(AppDisplayName, 10),
FirstSeen = min(TimeGenerated), LastSeen = max(TimeGenerated)
by UserPrincipalName, IPAddress
| where Failures >= 5
| order by Failures desc;
// 2. Successful admin sign ins that needed only one factor
SigninLogs
| where TimeGenerated > ago(Lookback)
| where UserPrincipalName in~ (Admins)
| where ResultType == "0"
| where AuthenticationRequirement == "singleFactorAuthentication"
| project TimeGenerated, UserPrincipalName, IPAddress, AppDisplayName Both queries read the Entra sign in log through the SigninLogs table, where ResultType is "0" for success and any other value for a failure, and AuthenticationRequirement records the highest level of authentication the sign in needed. The second query should return nothing at all for a well run tenant. If it returns rows, that list is the work for this week. Our Entra sign in and audit log review covers the rest of the reading, including role changes and new authentication methods registered to admin accounts.
The third place to look is not a log at all. Ask the help desk whether anyone has called in the last few weeks asking for an MFA reset or a new phone on an admin account. A pretext that uses the tenant's own name and the admin's real address is exactly what that call would sound like. Our help desk identity verification checks describe how we test that process when a client asks us to.
How we use tenant recon on an engagement
On an identity engagement the first deliverable is rarely an exploit. It is a map. We work out who administers the tenant, what their accounts look like, whether the tenant is hybrid, and which sign in methods are allowed. From the outside that means public DNS and mail records, the public OpenID configuration, and whatever the organization has published about its own staff. With one ordinary account inside, an Entra tenant gives up much more, because by default an ordinary member can read almost the whole directory. Our Entra ID first hour walks through that read.
The Connect Health fields line up with that map almost one for one. That is why we read this incident as a recon event rather than a data loss event. In the scenarios we run, the map is what turns a generic phishing test into a targeted one, and targeted is where the controls get tested properly.
What we then test, with the client's written scope and an approved pretext, is the part this breach cannot do on its own: whether a named administrator can be moved off the real sign in page, whether the help desk will reset a privileged account on the strength of details that are now known to an outsider, and whether a captured session for an admin survives the response. We never take an admin session further than the scope letter allows, and the evidence we hand back is the path, not a trophy.
The pattern that holds up best in our experience is plain. Admin roles sit on separate, cloud native accounts that do not receive mail, with phishing resistant methods enforced by Conditional Access, and with standing Global Administrator assignments kept to a very small number. Microsoft's own role guidance says the same: use cloud native accounts for Entra roles, avoid on premises synced accounts for role assignments, and keep Global Administrators to fewer than five. A tenant built that way loses very little when its admin addresses leak, because the addresses lead to accounts that cannot be phished in the usual ways and do not read email.
The hybrid flag deserves its own line. A tenant that syncs from Active Directory inherits the security of that directory and of the sync server. We treat the Entra Connect server as a tier zero asset on every hybrid engagement, because it is one. An attacker who knows a tenant is hybrid, and knows sync ran recently, knows the on premises side is a real path into the cloud tenant. That is the case for testing who holds privileged roles and how they are activated on both sides of the sync, not just in the cloud.
What is still coming
Expect the targeting to come before the details. If whoever read the API had a use for the list, it will show up as phishing and help desk pressure aimed at administrators, quite possibly with Microsoft themed pretexts that use the breach itself as the lure: "your tenant was affected, sign in to review". The notice gives attackers a reason to write to admins that admins half expect to receive. Tell your administrators now that any follow up about this incident will arrive in the Microsoft admin center and Service Health, not as a link in an email.
Expect more detail from Microsoft, too, and read it closely when it comes. The useful facts are the ones still missing: the date of the first unauthorized read, the number of tenants, and whether the activity looked targeted or broad. Each of them changes how far back a customer should hunt. If Microsoft publishes a fuller write up or a post incident review, we will update this article with what it says.
And expect this to be a reminder of a broader point about cloud services. Customers cannot patch them, scan them, or see their bugs until the provider says so. The protection is a tenant that assumes its administrators are known: few of them, on separate accounts, signing in only with methods that a proxy cannot relay. That design was right before 9 Nov. This incident just made the case for it in the provider's own words.
What to do today
-
Enforce phishing resistant sign in on every privileged role. Use a Conditional Access policy with the built in phishing resistant authentication strength: Windows Hello for Business, FIDO2 keys or certificate based authentication.
-
Treat admin addresses as known. Brief every administrator and the help desk that targeted phishing and reset requests are likely, and that follow ups about this incident arrive in Service Health, not by email link.
-
Run the two KQL queries above against your admin accounts, as far back toward 23 Jul as your retention allows, and fix every single factor admin sign in the second one finds.
-
Review Connect Health itself. Run Microsoft's reference scripts, check who holds Owner, Contributor and Reader, and confirm where alert mail goes.
-
Shrink the target. Move admin roles to cloud native accounts with no mailbox, keep Global Administrators to fewer than five, and keep two cloud only emergency access accounts.
-
Treat the Entra Connect server as tier zero in a hybrid tenant, with the same access controls and monitoring as a domain controller.
SOURCES
- BleepingComputer // Microsoft's notice: read only access to tenant metadata including admin and technical contact emails, fix deployed 19 Nov 2026
- Microsoft Learn // Connect Health monitors on premises identity infrastructure through agents on Entra Connect Sync, AD FS and AD DS servers
- Microsoft Learn // Connect Health roles, and alert mail that can go to all Global Administrators and custom addresses
- Microsoft on GitHub // read only reference scripts that show what the Connect Health APIs return for a tenant
- Microsoft Learn // role best practices: cloud native admin accounts and fewer than five Global Administrators