AegisAI Perspectives · 077

How do I spot a hijacked mailbox when MFA didn't stop the attacker?

Look after the login: a stolen session passes MFA, so check sign-in anomalies first, then hidden inbox rules, then unusual mail sent from the account.

How do I spot a hijacked mailbox when MFA didn't stop the attacker?

Look after the login, because a stolen session token works even when the owner satisfied multifactor authentication1. AegisAI reads sign-ins and mailbox activity per account and raises one explained finding when behavior stops matching the owner20. The checks below work in Microsoft 365 with or without it.

What you can do by the end

  • Check sign-in records for the session theft signs that remain after MFA was satisfied.
  • Find hidden inbox rules and forwarding that an attacker sets up inside a hijacked mailbox.
  • Contain the account in the right order, so a password reset does not leave the attacker inside.

Why does MFA not stop a hijacked mailbox?

MFA checks the person at sign-in, and the attacker steals what comes after it. A token theft attack replays tokens issued to a user, even if that user has satisfied multifactor authentication. Because authentication requirements are met, the attacker is granted access to organizational resources with the stolen token2.

The common route is a phishing proxy. Microsoft's Adversary in the Middle detection fires when an authentication session is linked to a malicious reverse proxy3. In that attack the adversary can intercept the user's credentials, including tokens issued to the user4. The page on how account takeover works explains the attack itself; this page is the checklist for spotting it.

Check the sign-in records first

Start with sign-ins, because the stolen session has to be used from somewhere. Entra ID Protection flags an anomalous token when the token shows an unusual lifetime or is played from an unfamiliar location5. It flags atypical travel when two sign-ins come from distant locations and one of them is atypical for the user6.

Look at the device and the network as well. Unfamiliar sign-in properties include the IP address and the device, among others7. A finance user who always signs in from one laptop and suddenly appears on a new browser from a hosting network deserves a closer look. Filter for non-interactive sign-ins too: Microsoft says an unfamiliar properties detection on a non-interactive sign-in deserves increased scrutiny due to the risk of token replay attacks24. Those are the background sign-ins an app makes with a refresh token, which is exactly what a stolen session uses.

Expect gaps for new staff. New users sit in a learning mode for unfamiliar sign-in properties, and the minimum duration is five days8. An account created this week can be hijacked before the detection has a baseline.

Check which detections your license shows

Check your license before you trust an empty report. Full details on most risk detections require Microsoft Entra ID P211. On a lower plan, read the raw sign-in logs and the mailbox rules by hand instead.

Look for hidden inbox rules and forwarding

Search the mailbox rules next, because a hijacked mailbox is usually set up to hide replies. Microsoft lists rules that automatically forward email to unknown addresses as a symptom of a compromised mailbox12. Entra ID Protection raises a detection when suspicious rules that delete or move messages or folders are set on a user's inbox9.

A forwarding rule that copies all mail to an outside address is its own detection10. Run Get-InboxRule with the IncludeHidden switch, which reviews all Inbox rules, including hidden ones, for rules that redirect or forward messages13. A rule named with a single dot that moves mail about invoices to a little used folder is a strong sign.

Account changes count too. Frequent password changes or unexplained account lockouts can indicate a compromised mailbox14. So can a mailbox that Microsoft 365 has blocked from sending email25, which usually means the account already sent a burst of spam or phishing. Treat a send block on a finance or executive account as a hijack until the sign-in records prove otherwise.

Read the mail the account sends

Read the sent mail last, because the hijacked account now phishes from inside. That mail passes every sender check, since it really is the account sending21. The text is often polished: the FBI says criminals use AI-generated text to appear believable to a reader in spear phishing schemes18.

Watch for a burst of internal mail asking colleagues to sign in, or a reply in a real invoice thread with new bank details. Business email compromise cost $3,046,598,558 in 2025, per the FBI IC3 annual report17.

Follow these steps when one account looks wrong

  1. Pull the account's sign-ins for the past week and mark every new device or location.

  2. Check Entra ID Protection for anomalous token or Adversary in the Middle detections on the account.

  3. Run Get-InboxRule with IncludeHidden and record every rule that forwards, moves or deletes mail.

  4. Search the account's sent mail for messages to many colleagues or to finance contacts.

  5. Revoke all sessions with Revoke-MgUserSignInSession, which also invalidates existing refresh tokens16.

  6. Reset the password and delete app passwords, because they aren't automatically revoked when you reset the password15.

  7. Remove the bad rules, then warn anyone who received mail from the account during the window.

Map each sign to where you check it

SignWhere to checkLicense note
Token played from an unfamiliar locationEntra ID Protection, anomalous tokenFull details need Entra ID P2
Session linked to a phishing proxyEntra ID Protection, Adversary in the MiddleFull details need Entra ID P2
Sign-ins from distant places close togetherEntra ID Protection, atypical travelRaw sign-in logs on lower plans
Hidden rule moving or deleting mailGet-InboxRule with IncludeHiddenAny Exchange Online admin
Forwarding to an outside addressGet-InboxRule, forwarding detectionAny Exchange Online admin
Unusual mail sent from the accountSent items and message traceAny Exchange Online admin

Illustrative example, figures are fictional. A controller at a 600 person firm approves a sign-in prompt on a fake document page at 09:12. At 09:31 the account signs in from a new browser on a hosting network, and at 09:36 a rule named "." starts moving mail containing "invoice" to an archive folder. The admin finds the rule with Get-InboxRule, revokes all sessions, deletes two app passwords and recalls a payment change sent to 14 suppliers.

How AegisAI handles this

A phishing proxy keeps the session cookie, so no second factor is asked for19. AegisAI agents read sign-ins and mailbox activity per account, and raise one explained finding when behavior stops matching the owner. Every finding comes with recommended steps, and your team acts in Entra ID, Okta or Google Workspace22. Early access runs on the connection you already have, review-only, with no MX change23.

How to measure it in 30 days

Track three numbers for the month after you start these checks.

  • Accounts with an anomalous token, proxy or atypical travel detection, and how many you confirmed as hijacked.
  • Hidden or forwarding inbox rules found by Get-InboxRule, and how many no user could explain.
  • Minutes from the first odd sign-in to revoked sessions on each confirmed account. The account takeover page shows one finding per account, and a trial on your own tenant gives a first baseline.

Questions people ask

How can a mailbox be hijacked if MFA is on?

The attacker steals the session after sign-in. A token theft attack replays tokens issued to a user even after MFA was satisfied, so the attacker never faces the second factor.

What is the first sign of a hijacked Microsoft 365 mailbox?

Often a sign-in from a new device or location shortly after the owner's own login. Entra ID Protection flags tokens played from an unfamiliar location. A new inbox rule soon after is a stronger sign.

Does resetting the password remove the attacker?

Not on its own. Revoke all sessions as well, and delete app passwords, because they aren't automatically revoked when you reset the password.

Can I spot this without Entra ID P2?

Partly. Full details on most risk detections need P2, but any Exchange Online admin can run Get-InboxRule with IncludeHidden and read the account's sent mail.

Why do colleagues trust mail from a hijacked account?

It really is the colleague's account, so every sender check passes. AI-generated text also makes the request read naturally.

Sources

  1. Microsoft Learn, Token theft playbook
  2. Microsoft Learn, Entra ID Protection risk detections
  3. Microsoft Learn, Respond to a compromised email account
  4. FBI IC3, Public Service Announcement I-120324-PSA
  5. FBI IC3, 2025 Internet Crime Report
  6. AegisAI, Account Takeover Protection

Recent posts

Every email, investigated.