in this guide
  1. what the vendors actually ask for
  2. why it becomes the most dangerous account you have
  3. the MFA change that caught the old setup
  4. setting it up so it holds on both days
  5. the drill nobody schedules
  6. beyond the identity provider

Every admin login you have depends on something that can break. The identity provider you federate with. The phone that receives the code. The Conditional Access policy someone tightened on a Friday. The one person who held Global Administrator and has now left. On a normal day none of that matters. On the day it all fails at once, you need one account that still opens the door.

That account is called break-glass, or emergency access. Microsoft and Google both tell you to have one. Microsoft’s documentation says to use it only for “break glass” scenarios where normal administrative accounts cannot be used.

The trouble is in how it is built. To survive the day everything fails, the account has to be exempt from the things that fail. So it skips the policies, belongs to nobody, and holds the highest role permanently. Described that way, it is the account an attacker would design for themselves.

This guide is about keeping both properties: an account that works on the worst day and is watched on every other day. The founder departure guide makes the case for having one at all. This guide is about running it.

The cloud account above every admin you have applies the same idea to AWS, Google Cloud and Azure.

what the vendors actually ask for

Microsoft’s guidance on emergency access accounts is specific, and as of November 2026 it reads like this:

  • Two or more accounts, cloud-only, on the tenant’s own onmicrosoft.com domain, never federated or synced from on-premises Active Directory.
  • Global Administrator, permanently active. If you use Privileged Identity Management, the emergency accounts are the exception: no activation, no approval.
  • Excluded from any Conditional Access policy that blocks or restricts sign-in, ideally through one dedicated group.
  • Not tied to any person and never registered to an employee’s phone. Credentials sit in secure, fireproof safes in separate locations.
  • A phishing-resistant sign-in method, with a passkey on a FIDO2 security key as the recommended option and certificate-based authentication as the alternative, and different from what your normal admins use.
  • An alert on every sign-in, a review after every use, and a test at least every 90 days.

Microsoft’s best-practices page for Entra roles adds that fewer than five people should hold Global Administrator, and that two cloud-only emergency accounts should hold it permanently. CISA’s secure configuration baseline for Entra ID (MS.AAD.7.1v1) sets a minimum of two and a maximum of eight Global Administrators, and its rule against permanent active admin assignments (MS.AAD.7.4v1) names emergency access accounts among its exceptions.

Google’s admin pages cited here do not use the words break-glass or emergency access. Its guidance for Workspace says an organisation should have more than one super admin account, each managed by a separate person, kept apart from the account used for daily work, protected by a security key, with a backup key and printed backup codes stored somewhere safe. Google Cloud’s super admin page adds that the security key for the first super admin should be kept safe, preferably at your physical location. Read together, that is the same idea in different words.

why it becomes the most dangerous account you have

Put the vendor checklist next to what usually exists in a small tenant and the gap is plain.

what the emergency account skips, and what still guards it

exempt by design

so it works on the bad day

  • Conditional Access block and restrict policies
  • Federation and on-premises sync
  • Role activation and approval
  • Any one person's phone or laptop

the only things left

so it is safe on every other day

  • The credential itself, ideally a hardware key in a safe
  • An alert that fires on every sign-in
  • Someone who reads that alert
  • A review after each use

Picture a 90-person firm where the first IT contractor created an account called admin-emergency in 2021. It got a long password, which went into a shared vault. It was excluded from the MFA policy so it could never lock anyone out. Nobody has signed into it since. Nobody set an alert, because the alert guide needed a log workspace the firm never built.

That account has every property on the left of the diagram and none on the right. It is a permanent Global Administrator, protected by a password that several people and one former contractor have seen, exempt from the policy that would stop a password spray, and silent when used.

An emergency account nobody watches is a back door with good paperwork.

The pattern is familiar from larger organisations. Microsoft’s own report on the Midnight Blizzard intrusion, published on 25 January 2024, says the attackers used password spraying to compromise a legacy, non-production test tenant account that did not have MFA enabled. That was a test account. What it shares with a neglected emergency account is the dangerous part: an account kept outside the normal rules, for a reason that made sense when it was created, and not looked at since.

the MFA change that caught the old setup

Microsoft’s page on mandatory MFA says break-glass and emergency access accounts must now sign in with MFA to reach the Azure portal, the Microsoft Entra admin center and the Microsoft 365 admin center, whatever your own Conditional Access exclusions say, and recommends a passkey on a FIDO2 key or certificate-based authentication for them. So a password-only emergency account, excluded from your MFA policy, meets a prompt it cannot answer on the admin portals it was created to reach, and nobody finds out until the day it is needed unless someone tests it. Google’s 2-Step Verification enforcement for admins points the same way, and the vendor login changes guide has the phases and dates for both.

setting it up so it holds on both days

The vendor guidance turns into a short list of decisions for a company of 20 to 500 people.

Two accounts, two keys each. Two accounts survive one being misconfigured. Two FIDO2 keys per account survive one key being lost. Keep them in two places, for example the office safe and a sealed envelope with a director. Microsoft’s page asks for separate locations and for safe combinations to change when someone with access leaves.

Nobody’s name on it. No personal phone number for recovery, no personal email as the alternate contact, no authenticator app on someone’s phone. If the account depends on a person, it fails on the day that person is the problem, the situation the founder departure guide is built around.

Excluded from the blocking policies, still covered by the strong one. Microsoft’s guidance is to exclude emergency accounts from Conditional Access policies that block or restrict sign-in. The passkey is what protects the account, and the passkeys guide covers where a passkey can quietly fall back to something weaker. Keep the exclusion group short, and check it on the same day you check the account, because a group anyone can add themselves to is a quiet way to escape every policy you have.

An alert that reaches a human. Microsoft’s documented method sends sign-in logs to Azure Monitor and fires a critical alert on any sign-in by either account. Whatever tool you use, the test is simple: does a sign-in by this account produce a message on a phone that someone answers, within minutes? In Google Workspace, admin email alerts and the admin log cover the same ground for super admin activity.

A rule for after. Every use is either a drill, a real emergency or misuse. Microsoft’s page asks for a review after each one, with the logs preserved. Write down the date, who used it and why, then replace the key or credential if it left the safe. The super admin on 12 March guide explains why that record has to live somewhere other than the system that produced it.

the drill nobody schedules

An emergency account that has never been used is a theory. Microsoft’s page asks for validation at least every 90 days, after any change in IT staff and when subscriptions change. The founder departure guide suggests once a year as a floor. For most small companies, the right answer is whatever you will actually put in a calendar, and a quarter is a sensible default.

a quarterly drill, in order

  1. 1tell the watchersWhoever gets the alert knows a test is coming.
  2. 2take the key from the safeTwo people, the documented procedure, nothing from memory.
  3. 3sign in and do one admin taskOn the admin portal, under today's policies.
  4. 4did the alert fire?If nobody was told within minutes, the account is unwatched.
  5. 5log it and put it backDate, who, result. Change the safe code if someone has left.

The drill catches the failures that only show up on the day: the key that was reset, the policy that now blocks the account, the alert that goes to a mailbox nobody reads, the account disabled by an inactivity clean-up. Microsoft’s page explicitly warns that the credential must not expire or fall into automated clean-up for lack of use. Switch it off and wait for the complaint makes the same point from the other side: emergency accounts are on its list of silent paths that must never be switched off as a test.

Two more things the drill should check, both from Microsoft’s validation list. That nobody has registered MFA or self-service password reset on the account to their own phone or personal details. And that the list of people allowed to open the safe still matches the people who work there.

beyond the identity provider

The identity provider is the obvious place for an emergency account, and most guidance stops there. Microsoft’s own page notes that most organisations need emergency access for other critical systems as well. For a small company, the short list is the systems where losing admin access stops the business:

  • the domain registrar and DNS, covered in the domain registrar guide
  • the device management console, where an approval scheme needs a documented way around it, as the device console guide explains
  • the cloud account root user, and the payment provider
  • the backup console, if it uses a separate login

Each of these needs the same three answers: who can get in if the usual admin cannot, what that login depends on, and when someone last proved it works. The account that reaches all of them is part of your blast radius, so it belongs on that list too.

When the alert from your test arrives, ask the person who received it what they would have done if it had not been a test. Do they know?

on the record

all myths →
read next in the platformsD12 the cloud account above every admin you haveeach cloud has one login that sits over the admin roles and can hand them back to itself. it was created on the first day by whoever signed up, and it still runs on their email and their phone.

sources

  1. Microsoft Learn: Manage emergency access admin accounts in Microsoft Entra ID
  2. Microsoft Learn: Plan for mandatory Microsoft Entra multifactor authentication
  3. Microsoft Learn: Best practices for Microsoft Entra roles
  4. CISA ScubaGear: Microsoft Entra ID secure configuration baseline (MS.AAD.7)
  5. Google Workspace Admin Help: Security best practices for administrator accounts
  6. Google Cloud: Super administrator account best practices
  7. Google Workspace Admin Help: About 2SV enforcement for admins
  8. Microsoft Threat Intelligence: Midnight Blizzard, guidance for responders on nation-state attack, 25 January 2024