Choosing Which Apps to Protect When Entra Is Down

Share

Thursday afternoon, the help desk lights up. Users can't get into the ticketing system to tell you they can't get into anything. Someone who's still signed in to Teams asks if "Azure is down."

That was November 12. Starting around 1:26 PM EST, Entra SSO sign-ins began failing, and for many organizations it stayed that way for the better part of eight hours. The pattern was telling: anyone who already had a session, or a local username and password, kept working. Anyone who needed a fresh sign-in through Entra was stuck.

Eight hours is a long time to not be able to open a door.

I'm not going to try to explain what happened on Microsoft's side. What I can talk about is the decision I keep seeing teams get wrong: "We should add a local backup login to everything."

Please don't.

Why "just add a backup login everywhere" is the wrong lesson

A local account feels like cheap insurance. It isn't. Every one of those accounts is another password to protect, another way around Conditional Access and MFA, and another object that will get missed when someone leaves the company. It's easy to picture: offboarding runs through Entra, and six months later a contractor's backup login is still sitting on some SaaS tool.

You also just taught your users a second way in. When the outage is over, some of them will keep using it.

SSO exists for a reason. The goal isn't to undo it. It's to decide, ahead of time, which doors still have to open if sign-in is down for four to eight hours — and to protect those doors on purpose.

A simple three-tier approach

I like a blunt test: if Entra can't issue or refresh a sign-in for the rest of the business day, what actually breaks? Ask the app owners questions that force a real answer, not a vibes-based "it's important."

  • If this app is down until tomorrow morning, does someone get hurt, or does a hospital, dispatch, or floor workflow freeze?
  • Do we stop taking revenue or serving customers? Can we take a ticket by phone and enter it later?
  • Who is stuck — a handful of admins, a whole department, or every customer on a portal?
  • Do we need every user, or just one or two people who can keep the lights on? Is there already a paper or email workaround?

Then put every Entra-federated app into one of three tiers.

Tier 1 — Safety, operations, and the business cannot pause. A safety issue, a regulatory problem, or a core operation that can't be deferred: clinical or dispatch systems, plant-floor tools, the phone system, the SOC console, payroll if you're on a pay run that day. This should be a short list. If everything is Tier 1, nothing is.

Tier 2 — Revenue, customer-facing, or an internal process that gets expensive if it stops. A customer portal, checkout, or the ticketing tool your support team lives in. A half-day outage is painful, but you can often degrade: take orders by phone, or have a manager complete what can't wait.

Tier 3 — Nice to have. The marketing CMS, the survey tool, that one SaaS app three people love and everyone else forgot about. It can wait. Spend your energy on Tiers 1 and 2.

What protection actually looks like

Vaulted break-glass accounts on the app. For Tier 1, and maybe the admin side of Tier 2, keep a local admin that doesn't depend on Entra. Store it in a vault, alert on any use, and limit it to the one or two operators who'd actually need to run the app. It can bypass Conditional Access and MFA, so treat it like a privileged credential and review it when people change roles.

Longer-lived sessions, used carefully. Users who already had a valid session often rode through this outage. That's your hint. In Entra, use Conditional Access session controls — sign-in frequency and persistent browser sessions — rather than inventing custom token lifetimes. Microsoft retired configurable lifetimes for refresh and session tokens in favor of those controls. A short sign-in frequency, especially "every time," means more trips back to Entra — rough when Entra can't issue a new token. A more relaxed frequency plus persistent browser sessions on managed devices helps people already at work keep working.

Continuous access evaluation (CAE) is easy to get backwards. For CAE-capable clients talking to Exchange Online, Teams, and SharePoint Online, Entra can issue access tokens that last up to 28 hours and revoke them near real time on critical events. That can help an already-established session later in the token's life, but it doesn't help a brand-new sign-in, and it isn't a setting you flip on for every third-party SAML app.

Keep critical apps signed in on shared or kiosk devices. If the warehouse scanner, front-desk kiosk, or SOC dashboard is already signed into a Tier 1 app at the start of the shift, an afternoon IdP outage doesn't brick the floor. The tradeoff is physical access: screen locks, badge control, and a rule about who walks away logged in.

A secondary IdP, for a few apps only. You can point a handful of Tier 1 apps at a second identity provider, or run a carefully designed Okta-to-Entra federation so an app can still authenticate if one hop is unhealthy. I'll tell you the quiet part: this is costly, messy, and easy to get wrong. Do it for the apps you truly can't afford to lose, not as a standard build.

Documented manual or offline procedures. Sometimes the fallback isn't a login at all. It's a paper runbook or a phone script for taking orders and keying them later. Unfashionable, extremely effective, and you didn't create another account.

And keep your Entra emergency access accounts. App-side fallbacks don't help if you can't get into the tenant. Microsoft's guidance is to maintain cloud-only emergency access ("break-glass") accounts: at least two, on the *.onmicrosoft.com domain, not federated or synced from on-prem, excluded from Conditional Access, and protected with a strong method such as a FIDO2 passkey. Test that you can sign in with them.

Write it down

If it isn't written down, you'll invent the plan during the incident. Use a table short enough that you'll actually maintain it:

App Owner Tier Fallback method Where the creds live Last tested
Plant floor MES Operations 1 Vaulted local operator account; kiosk stays signed in Privileged vault 2026-10-15
Customer support portal Support Director 2 Manager local admin; phone-and-queue runbook Vault; outage runbook 2026-09-02
Employee recognition app HR 3 None, accept downtime n/a n/a

"Where the creds live" should be a vault path or "no creds, manual process," never a person's name.

Test it, or you don't have it

Run a tabletop: Entra sign-in is unavailable for eight hours, starting now. Who is stuck? Which Tier 1 apps still open? Who has the vault? How do we tell the company if Teams and Outlook are in the blast radius too? You'll find the missing owner and the expired local password in that room.

Then, once a quarter, actually sign in with the fallback. Microsoft's emergency-access guidance is to validate those tenant accounts at least every 90 days, and the same cadence is a good habit for app-side logins. Update the "Last tested" column. If nobody will attest to it, it isn't a control.

Final thoughts

I like SSO. I like Conditional Access. I like not having a unique password for every portal in the company. An Entra bad day doesn't mean that model is wrong. It means a handful of apps were more important than we treated them, and a much larger handful weren't important enough to earn an exception.

Tier the estate. Protect the doors that still have to open. Leave the rest federated.

Stay curious. Ask the app owner what happens if nobody can sign in until tomorrow. And maybe don't create a local admin named backup on the marketing CMS at 2 AM just to feel like you did something.