An account with MFA on can be compromised with MFA on. The attacker lures the user to a fake sign-in page and relays the password and one-time code to the real service in real time; the second factor is verified and the session cookie ends up in the attacker’s hands. How well do the MFA methods you have hold up in this scenario, how do you choose between a FIDO2 security key, Windows Hello for Business and passkeys, and how do you move to phishing-resistant MFA in Microsoft Entra ID without locking users out — we look at these in order.
Which MFA method falls to which attack
The answer “we have MFA” doesn’t say which method is used. The decisive question is: can an attacker trick the user and use the second factor in their own session?
| Method | How it holds up against attacks | Phishing-resistant? |
|---|---|---|
| SMS / voice call | SIM swapping, the code entered on a fake page | No |
| TOTP (app code) | The code is entered on a fake page and relayed via AiTM | No |
| Push notification | MFA fatigue, approval during AiTM | No (number matching reduces the risk) |
| FIDO2 / passkey | An origin-bound signature; doesn’t work on a fake domain | Yes |
| Windows Hello for Business | A TPM-bound key that never leaves the device | Yes |
| Certificate-based authentication | The private key is on a smart card or the device | Yes (when configured as multifactor) |
In an adversary-in-the-middle (AiTM) attack, the fake page works like a reverse proxy between the user and the real sign-in service. The user enters their password and code, the real service completes MFA successfully, and the attacker copies the returned session cookie. After that, they need neither the password nor the code. MFA fatigue works differently: the attacker already has the password, sends push requests one after another and waits for the user to tap “Approve” on one of them. Number matching in Microsoft Authenticator makes this second scenario significantly harder. It doesn’t stop AiTM. Because the user believes they’re signing in for real, they enter the number on the screen with their own hand.
A common picture in incident investigations: the sign-in log shows MFA as “successful”, and the user says “I entered the code myself.” Both are true. The problem is the method itself.
Phishing resistance comes from the protocol
FIDO2 consists of the W3C WebAuthn and FIDO Alliance CTAP protocols. During registration, the authenticator (a security key, phone or TPM) generates a separate key pair for each service; the private key never leaves the authenticator, and only the public key is registered with the service. During sign-in, the service sends a challenge and the authenticator signs it. The crux is that the signature is bound to the relying party ID (RP ID), i.e. the domain. The browser itself passes the origin the request came from to the authenticator; a credential registered for login.microsoftonline.com is never offered on a lookalike fake domain. Even if the user doesn’t notice the fake page, a signature can’t be produced.
In other words, protection doesn’t depend on the user’s attention. The protocol does it itself.
Know two limits from the start. First: malware on the device stealing session tokens after authentication is complete is outside FIDO2’s scope; that requires endpoint security, a compliant device condition and session controls such as token protection. Second is the registration and recovery path: if a user can register their new key with a weak method, an attacker can add their own key through the same door.
FIDO2 keys, Windows Hello and passkeys: which for whom
The three options share the same cryptographic foundation; deployment, cost and daily use differ. In enterprise cybersecurity designs, building a mix by user profile instead of a single method is usually more realistic.
| Criterion | FIDO2 security key | Windows Hello for Business | Passkey in Authenticator |
|---|---|---|---|
| Where the key lives | A hardware key (USB/NFC) | The TPM on the device | Phone (device-bound) |
| Additional hardware | Required | Not required (TPM 2.0 recommended) | A corporate or personal phone |
| Shared device | Suitable | Not suitable | Partly |
| Signing in on another device | By plugging in the key | Only on the registered device | Cross-device via Bluetooth proximity |
| Typical user | Administrators, production, shift workers | Corporate laptop users | Mobile-first users |
A FIDO2 security key is the clearest option for administrators and emergency accounts: it’s device-independent, revoked when lost and can’t be copied. In Entra ID, key models can be restricted by AAGUID; that way only certified models purchased by the organization get registered. A cheap key a user bought themselves doesn’t sneak into the inventory.
Windows Hello for Business is the lowest-friction option for users; the PIN or biometric only unlocks the TPM key on the device and never travels over the network. In hybrid environments, the deployment model Microsoft recommends for single sign-on access to on-premises resources is cloud Kerberos trust; the key trust and certificate trust models bring additional PKI or domain controller requirements. This is where we often see people get stuck: certificate trust is chosen by following an old deployment guide, and the PKI burden is taken on unnecessarily.
Passkey is the general name for a FIDO2 credential. A device-bound passkey in Microsoft Authenticator turns the phone into a phishing-resistant authenticator without buying extra hardware. Support for passkeys synchronized with a cloud account is changing in Entra ID; verify against Microsoft’s current documentation before including them in policy, and decide whether the key leaving for a personal cloud account fits your risk model.
Authentication strength policy in Entra ID
Enabling a method doesn’t make it mandatory. Entra ID has two separate layers. The authentication methods policy determines which method is open for registration and use for which group. Authentication strength, within Conditional Access, defines which method combinations are accepted for a given access. There are three built-in strengths: Multifactor authentication strength, Passwordless MFA strength and Phishing-resistant MFA strength. The last accepts Windows Hello for Business, FIDO2/passkeys and multifactor certificate-based authentication. One is useless without the other.
- In the Entra admin center, under Authentication methods > Policies, enable the Passkey (FIDO2) method for the target group; if needed, enable the attestation requirement and AAGUID restriction.
- Open Temporary Access Pass (TAP) only to the registration and recovery group.
- Create a new Conditional Access policy: target users the pilot group, target resource all resources or admin portals first, access control Require authentication strength > Phishing-resistant MFA.
- Run the policy in Report-only mode first; watch for users who can’t meet the condition in the sign-in logs.
- Exclude emergency accounts from the policy, but protect these accounts with FIDO2 keys too.
For the general design of the policy framework, exclusion and naming rules, see the Conditional Access planning guide. The full method list for the built-in strengths is in Microsoft’s authentication strengths documentation.
Before turning on the requirement, measure the current registration state. Who’s ready and who isn’t — a report, not a guess. With Microsoft Graph PowerShell, registered methods per user are extracted like this:
Connect-MgGraph -Scopes "AuditLog.Read.All","UserAuthenticationMethod.Read.All"
# Registered methods per user
$report = Get-MgReportAuthenticationMethodUserRegistrationDetail -All
$report | Select-Object UserPrincipalName, IsAdmin,
@{n='Methods'; e={$_.MethodsRegistered -join ','}} |
Export-Csv .\mfa-registration-status.csv -NoTypeInformation -Encoding UTF8
# Administrators without a passkey or WHfB registration
$report | Where-Object { $_.IsAdmin -and
-not ($_.MethodsRegistered -match 'passKeyDeviceBound|fido2|windowsHelloForBusiness') } |
Select-Object UserPrincipalName
The value names returned in the MethodsRegistered field can differ by API version; adjust the filter to the sample output you get from your own tenant. If the filter comes back empty on the first run, look at the raw output first; the problem is usually how the value name is spelled.
Registration, lost keys and recovery
A phishing-resistant method is only as strong as the weakest registration path. If a user can register a new passkey from a session verified only by SMS, an attacker with a stolen SMS code can add their own key too. That’s why registration and recovery are designed as a separate process:
- Initial registration: A user whose identity has been verified by the help desk is given a short-lived, one-time TAP; the user signs in with the TAP and registers a passkey or WHfB. TAP lifetime is configurable; keep it short for registration.
- Identity verification: The verification step the help desk performs before issuing a TAP should be written down (e.g. a call-back from the manager, video verification). Social engineering targets exactly this step.
- A second authenticator: Register two security keys for administrators, with one kept as a backup in a secure place; this reduces dependency on the help desk when a key is lost.
- Lost key: The lost key is deleted from the user’s registered methods, active sessions are revoked (
Revoke-MgUserSignInSession) and re-registration is done with a TAP. - Registration policy: A separate Conditional Access policy for the “Register security information” user action can require registration to be done only from a trusted location, a compliant device or with a TAP.
Phased migration: start with administrators
The migration starts with the accounts where risk is highest; each phase is planned with support data from the previous one. Putting everyone under the requirement on the same day swamps the help desk on Monday morning.
| Stage | Scope | Requirement | Exit criterion |
|---|---|---|---|
| 0. Preparation | Methods policy, TAP, registration policy | None | The help desk procedure is ready |
| 1. Administrators | Entra roles, emergency accounts | Admin portals and all resources | 2 authenticators for every administrator |
| 2. Sensitive roles | Finance, HR, senior management, assistants | All resources | No failed sign-ins left in report-only |
| 3. General users | Department-based waves | All resources | Weak methods removed |
| 4. Exceptions | Shared devices, legacy applications | Compensating control | The exception list has owners and expiry dates |
For administrator accounts, handle the migration together with just-in-time authorization: requiring a phishing-resistant method at role activation, combined with Entra ID PIM configuration, closes the risks of standing privilege and weak MFA together. At the finance and senior management phase, also review phishing controls on the email side; this topic is described in the reducing phishing risk in enterprise email article.
Common mistakes:
- Turning on the authentication strength policy directly in “On” mode for all users and skipping the report-only phase.
- Not inventorying applications that sign in with legacy protocols (IMAP, POP, SMTP AUTH); these can’t use FIDO2 and must be blocked separately.
- Planning WHfB for shared production terminals; security keys are more suitable for this scenario.
- Continuing to protect the registration process with SMS.
- Counting the migration as “done” before risk-based policies are in place; for anomalous sign-ins, Entra ID Protection risk policies are a complement.
For measurement, watch the Authentication details field in the sign-in logs: in groups under the requirement, all successful sign-ins should be made with a phishing-resistant method, there should be no sign-ins falling back to a weak method, and the exception list should be reviewed every quarter. If the exception list has no owner, it grows. If you want to handle authentication design together with access processes as a whole, see our identity and access management service; controls for the Microsoft 365 tenant as a whole are covered under Microsoft 365 security.
Frequently asked questions
Does push notification with number matching count as phishing-resistant?
No. Number matching makes MFA fatigue attacks harder, but in an AiTM attack the user enters the number themselves, believing they’re signing in for real, and the session cookie is still stolen. Phishing resistance requires the credential to be bound to the domain; FIDO2, Windows Hello for Business and certificate-based authentication provide this.
What’s the difference between a passkey and a FIDO2 security key?
Both are FIDO2 credentials. A security key is a separate piece of hardware, and the credential never leaves it. A passkey is a credential stored on a device such as a phone or computer; it can be device-bound or synchronized with a cloud account. Decide separately in the organization’s policy which type will be accepted.
What should be done if a user loses their security key?
The lost key is deleted from the user’s registered methods, active sessions are revoked, and the user, once their identity is verified by the help desk, is given a short-lived Temporary Access Pass and has them register a new authenticator. Registering a second backup key for administrators from the start shortens this process.
Should emergency accounts use FIDO2 too?
Yes. Emergency accounts are usually excluded from Conditional Access policies; that’s why their own authentication must be strong. FIDO2 keys stored separately in a secure place both reduce the risk of the policy exclusion and meet the MFA requirement Microsoft has introduced for its admin portals.