AiTM Token Theft in Microsoft 365: How to Spot and Stop It

Microsoft 365 and CopilotPublished Last reviewed November 20269 min read

Short answer

AiTM token theft happens when a phishing proxy relays a real Microsoft 365 sign in, MFA included, and keeps the session cookie the user earned. Push approvals and codes cannot stop it because the user completes them. Respond by blocking sign in and revoking sessions, resetting the password, removing rogue MFA methods, inbox rules, forwarding and app consents, then require phishing resistant MFA and managed devices.

Worried about an account right now? Run the account takeover check, or call (786) 577-9743.

Update, November 2026

Microsoft now lists token protection as generally available for native apps on iOS, iPadOS and macOS as well as Windows, while browser support is limited to a preview for selected apps that access Azure Resource Manager, so browser sessions to Outlook and SharePoint are still not covered. Microsoft also later set July 1, 2027 as the SMS and voice retirement date for Global Administrators and external users; February 1, 2027 still applies to everyone else.

Adversary in the middle (AiTM) phishing does not break multifactor authentication. It walks your user through a real Microsoft 365 sign in, MFA included, and keeps the session that sign in produces. If your tenant relies on push approvals or codes, the attacker ends up holding a valid session, and a password reset alone may not end it right away. The response has an order, and the lasting fix is a sign in that cannot be relayed.

What is AiTM phishing and session token theft in Microsoft 365?

AiTM phishing puts a reverse proxy between the user and the real Microsoft sign in page. The user signs in to Microsoft through the attacker’s server, and when Microsoft Entra ID issues the session cookie at the end, the proxy keeps a copy and replays it from the attacker’s own browser.

Microsoft documented this pattern in 2022 in a campaign that attempted to reach more than 10,000 organizations. Within minutes of stealing a cookie, the attackers opened Outlook on the web, searched for payment and invoice threads, and created inbox rules that hid replies from the real user. That is still the pattern to expect: the target is the mailbox and the money discussed in it.

Why doesn’t push or code MFA stop AiTM attacks?

Because the real user completes the MFA prompt against Microsoft’s real service, and the token that proves it is what gets stolen. Push approval, number matching, SMS and authenticator codes all confirm that the right person is present. None of them can tell that the page in front of that person is a proxy.

Microsoft’s token theft playbook puts it plainly: the attacker replays a token issued to a user who already satisfied MFA, so access is granted. What breaks the relay is a method bound to the real site. A passkey (FIDO2) or Windows Hello for Business signs a challenge tied to the domain it was registered with, so a lookalike domain receives nothing it can reuse.

What are the signs of AiTM token theft in Entra sign in logs and Entra ID Protection?

The clearest sign is one session used from two places: a successful MFA sign in from an address the user never uses, then activity from somewhere else carrying the same Session ID. Entra ID Protection adds named detections, depending on your license.

SignalWhere to lookWhat it suggests
Successful MFA sign in from an unfamiliar IP address or hosting provider, shortly after a link was clickedEntra sign in logs, Location and Authentication Details tabsThe proxy completed the sign in for the user
The same Session ID from a second IP address or device, with MFA satisfied by a claim in the tokenEntra sign in logs, Basic info tabThe stolen session is being replayed
Attacker in the Middle (user raised to High) or Anomalous tokenEntra ID Protection risk detections and risky usersMicrosoft linked the session to a malicious proxy or saw token replay
A new authentication method the user did not addEntra audit logs, User registered security infoPersistence, so the attacker can sign in again later
New inbox rules that move, mark read or delete mail, or new forwardingExchange audit records, Suspicious inbox manipulation rules detectionBusiness email compromise in progress
Consent to application by the userAudit log and the user’s applications in EntraAn OAuth app that keeps access after a password reset

License matters here. On Microsoft 365 Business Premium, which includes Entra ID P1, the premium risk detections show up only as Additional risk detected. Anomalous token and Unfamiliar sign in properties need Entra ID P2, and Attacker in the Middle and the inbox rule detections also rely on Microsoft Defender for Cloud Apps. Without those licenses, the sign in logs and audit logs are your detection, so someone has to know how to read them.

What should a small or midsize firm do first after AiTM token theft?

Contain the account first, and remove what the attacker added before you work out what they touched. The mistake to avoid is revoking sessions and stopping there, because the attacker signs back in through whatever they left behind. Ethereum Tech has run a full Microsoft 365 incident response for an AiTM session token theft, and this is the order we work in.

  1. Block sign in and revoke sessions. In the Microsoft Entra admin center, turn off Account enabled, then select Revoke sessions. Access tokens last an hour by default, but Microsoft notes that Exchange Online, SharePoint Online and Teams use continuous access evaluation to act on revocation in near real time.
  2. Reset the password and remove MFA methods you did not issue, in the same sitting. An attacker who registered their own authenticator app or phone number may be able to complete MFA, or even self service password reset, after you change the password. For accounts synced from on premises Active Directory, Microsoft recommends resetting the password twice.
  3. Check inbox rules, including hidden ones. Look for rules that move mail to Archive, RSS Subscriptions or Junk Email, mark it read, or delete it. These rules hide replies from the real user while the attacker works a payment thread.
  4. Check forwarding: forwarding addresses set on the mailbox, rules that forward or redirect, and new mailbox delegates. Then confirm your outbound spam policy still blocks automatic forwarding to outside addresses.
  5. Review OAuth app consents. Search the audit log for Consent to application during the attack window and review the user’s applications in Entra. A malicious app keeps its access after a password reset, so remove its permission grant.
  6. Remove devices the user did not register and roll back any role or policy changes made from the account.
  7. Scope the damage with the mailbox audit. Use the Session ID from the sign in logs to separate the attacker’s activity from the user’s, then search the audit log for MailItemsAccessed records and SharePoint and OneDrive downloads. Audit (Standard) keeps records for 180 days. If payment threads were touched, warn the other side by phone.
  8. Close the Conditional Access gap before you turn the account back on. Ask which policy should have stopped the replay. Common answers are no device requirement for browser access, an admin excluded from MFA, or an old exclusion nobody removed.

Keep notes as you go: timestamps, IP addresses, rule names and app IDs. Your insurer and bank may ask for them, and that record is where our cyber insurance evidence work starts. Microsoft’s guide to responding to a compromised email account is a good companion checklist.

How do you prevent AiTM token theft in Microsoft 365?

Require a sign in that cannot be relayed and a device the attacker does not have. Phishing resistant MFA defeats the proxy, and a device compliance requirement in Conditional Access refuses the attacker’s own machine.

Phishing resistant MFA, admins first

Register passkeys (FIDO2) on security keys or in Microsoft Authenticator, or set up Windows Hello for Business, then require the built in Phishing resistant MFA authentication strength in Conditional Access, starting with admin roles and excluding your emergency access accounts. On July 13, 2026, Microsoft announced that passkeys become the default sign in experience in Entra ID from September 1, 2026 and that Microsoft provided SMS and voice MFA retires on February 1, 2027, so this work has a deadline either way.

A device compliance requirement

With Entra ID P1 and Intune, require a device marked as meeting your Intune compliance policy for Microsoft 365 access. Build the compliance policies first, run the Conditional Access policy in report only mode, and block unknown device platforms so an attacker cannot claim one you do not check.

Token protection, where it applies

As of July 2026, the Conditional Access session control Require token protection for sign in sessions is generally available for native Windows apps that connect to Exchange Online, SharePoint Online and Teams, with macOS, iOS and iPadOS in preview on managed devices. It needs Entra ID P1 and registered or joined devices. Token protection does not cover browser sessions to Outlook on the web or SharePoint, where AiTM kits operate, so treat it as protection for tokens on devices, not an AiTM fix.

These controls also lift your Microsoft Secure Score, but choose them for the risk they remove, the same way we rank Secure Score fixes.

Is Microsoft 365 Business Premium enough to stop AiTM?

For prevention, mostly yes. Business Premium includes Entra ID P1 and Intune, which covers passkeys, authentication strengths, device compliance requirements and token protection. It lacks the P2 detection layer, so the gap is visibility, not blocking. A Microsoft 365 security assessment checks these exact settings against how your people actually sign in and work.

If you suspect a session was stolen, or you want these gaps closed before it happens, book a 30 minute call and we will talk through what to check first in your tenant.

Frequently asked questions

Can AiTM phishing steal a passkey?

No. A passkey signs a challenge bound to the domain it was registered for, so a proxy on a different domain cannot obtain a usable signature. Attackers may instead try to steer users to a weaker method, which is why the Conditional Access policy that requires phishing resistant MFA matters as much as registering the passkey.

How long does a stolen Microsoft 365 session stay useful to an attacker?

Until it expires or is revoked. Refresh tokens and session cookies can last far longer than the one hour default for access tokens, so waiting it out is not a plan. Revoking sessions, resetting the password and removing rogue MFA methods together are what end the attacker’s access.

Should we disable the user or only revoke their sessions?

Disable sign in while you clean up, if the business can absorb it. Microsoft lists disabling the account first because it stops new sign ins while you remove rules, methods and app consents. Revoking sessions alone leaves the account open to anyone holding the password or a rogue MFA method.

Are SMS codes any safer than push approvals against AiTM?

No. The user types the SMS code into the proxy, which passes it to Microsoft and keeps the resulting session, exactly as with push approvals or authenticator codes. SMS also carries SIM swap risk. Moving to passkeys or Windows Hello for Business removes both problems.

Who should investigate an AiTM incident at a small firm?

Someone with at least Security Reader or Global Reader access to Entra sign in and audit logs, plus Exchange Online permissions to review rules and forwarding. If nobody internal has done it before, bring in help early, because the evidence in audit logs ages out and every hour of delay gives the attacker more time in the mailbox.

Keep reading

Related guides

Have a question this raised?

Bring it to a 30 minute call and leave knowing where you stand and what comes next.

Book a 30 minute call

(786) 577-9743sales@ethereumtech.ai

Or take the two minute starting point check