Insights

AI Security

When M365 goes down, can you still log into your disaster recovery platform?

If the only second factor on your disaster recovery platform is a code sent by email, an email outage locks you out of the tool you need during the outage. The same outage silences any alert that travels only by email or Teams.

November 17, 2026 · 7 min read

When M365 goes down, can you still log into your disaster recovery platform?

A disaster recovery platform whose only second factor is a code sent to an Exchange Online mailbox cannot be reached while Exchange Online is down. That sounds obvious written out. Last week's Microsoft 365 outage made it concrete for anyone whose recovery tooling quietly depended on the system that was having the bad day.

According to Microsoft's preliminary post incident report, as reported by The Register, the cause was a defect introduced in July 2026 in Microsoft 365 network gateway servers. It passed load testing and sat quietly for months. Then two ordinary things happened together: Teams failed over onto the gateway that also serves Exchange Online, and almost 14 percent of gateway capacity was out of service. Neither was unusual alone. Together they set off the defect, and within two hours it had spread across the fleet. The report says that at the lowest point, "more than 80 percent of servers were unresponsive."

Two details matter for planning. Some services, including Teams, OneDrive, and SharePoint, were able to fail away from the faulty gateways. Exchange Online could not, for scale reasons, which is why Outlook was the part most people noticed. And a simple rollback was not possible because the change was months old, so engineers had to build an update that reverted the defective code path. A final post incident report is due November 20.

I am not going to add up how many people were locked out of what. Nobody has published that number, and the lesson does not need it. There are two lessons here, one about logins and one about alerts, and they share a cause.

Recovery tooling that depends on the system it recovers

Picture a team that did most things right. It has a disaster recovery platform, a runbook, and named people who know how to start a failover. Accounts on the platform require MFA. The MFA method, set at onboarding years ago, is a one-time code sent by email to each person's corporate mailbox.

Now take the mailbox away. Every login to the recovery platform stops at the second factor. The runbook is fine. The replicas are fine. The people are ready. The door is locked, and the key is in the building that is on fire.

Nothing in that setup looks wrong on a single review. The DR vendor offered email codes as a valid factor. The identity team saw MFA turned on and checked the box. The infrastructure team tested failover with everyone logged in through normal channels. Each review was correct about its own piece and blind to the path between them.

The rule I take from this: recovery tooling should not depend on the system it recovers. Email is the clearest case because so many login flows route through it, as a second factor, a password reset link, or an approval notice. The same rule applies to single sign-on. If every recovery tool authenticates through one identity provider, an outage at that provider becomes an outage of your recovery.

Alerts that travel only by email or Teams

The login problem has a mirror image. During an outage you need to reach your tools, and your tools need to reach you.

Think about where alerts can end up in a health system or HealthTech operation:

  • Critical lab values and imaging results routed to a shared mailbox for follow-up
  • Pharmacy and medication storage temperature alarms sent as email
  • Facilities alarms for generators, medical gas, or HVAC that notify a distribution list
  • Security tool alerts that land in a Teams channel the SOC watches
  • On-call escalations where the paging system sends to email or Teams as the last step

None of these is wrong on its own. The risk is an alert with exactly one path, where that path runs through the same vendor as everything else. During an outage, nobody receives the alert, and nobody knows they did not receive it. A silent channel looks the same as a quiet night.

Teams being able to fail away from the faulty gateways last week does not make it a safe backup for email. Both sit with the same vendor, and next time the failure may land on a different service.

One checklist for logins and alerts

You can run this in an afternoon without buying anything. Start with the systems you would need in the first hours of an incident: the DR platform and its admin console, backup software and its storage, the EHR's downtime procedures and wherever the downtime forms live, the password vault and whatever unlocks it, incident communications (paging tool, conference bridge, out-of-band chat), cloud consoles, and the identity provider's own admin portal. Then add every alert someone would need to act on within an hour.

For each system:

  1. How does a person sign in? Directly, or through SSO to an identity provider that also runs on M365?
  2. What is the second factor, and does any path to it touch email, Teams, or a phone number that forwards into M365? Remove email, and SMS that forwards into M365, as the only factor. An authenticator app or a hardware security key works when the mail service does not.
  3. How does a locked-out person recover access? If the answer is "a reset link to your work email," mark it.
  4. Is there a break-glass account outside SSO, with a non-email factor, stored offline in a sealed, logged location that two people open together? Alert loudly whenever one is used.
  5. When was that account last signed into? A break-glass account nobody has used in eighteen months is a theory. Sign in, confirm the key and the role, and reseal it.

For each alert:

  1. Write down every channel it travels through to reach a human: email, Teams, SMS, pager, phone call, a dashboard someone watches.
  2. Mark the alerts whose only paths depend on Microsoft 365, including SMS gateways and paging tools that deliver through Exchange.
  3. Give each of those a second path that does not share the dependency: a direct page, a phone tree, a local audible alarm, or a dashboard on a screen in the room.
  4. Add a heartbeat: a scheduled test alert that someone expects to see. If it stops arriving, you learn about the outage from your own process instead of from a missed result.

A tabletop inject that tests both

The inject is two sentences:

"Email and Teams are down across the organization. Start your DR runbook."

Then let the room work through it out loud, with IT, security, clinical operations, facilities, and pharmacy present. Who gets paged, and through what? Who opens the recovery platform first, and how do they sign in? What happens at the MFA prompt? Where is the break-glass account, who can open the envelope, and when was the key last tested? Which alerts have you not received since the inject started, and how would you know?

For more pressure, extend it: "M365 has been down for eight hours. Walk through getting into DR." The length forces two decisions. Do you wait for the vendor or fail over without them? And when service returns, how do you work through hours of delayed alerts so the oldest critical one is handled first instead of buried under the backlog?

A regular tabletop with that inject surfaces an email-only factor in the first ten minutes. The fix is a configuration change and a hardware key. Finding it during a real outage costs the hours you most needed.

What to do before the final report

Microsoft's final report on November 20 may change the details. It will not change the lesson. Your recovery plan has an authentication path and an alerting path, and both have dependencies that are rarely written down.

This week, run the checklist for your top five recovery systems and the ten alerts that would hurt most if missed. Fix the factor on the worst login first, give the worst single-path alert a second path and a heartbeat, then schedule the tabletop that proves both.