Guide Cybersecurity

How Attackers Get Around MFA: Phishing, Session Theft and Defences

Multi-factor authentication (MFA) significantly reduces the risk of account takeover through compromised passwords. Not every method offers the same protection against phishing, though. Adversary-in-the-middle (AiTM) proxies, session cookie theft, MFA fatigue and malicious app consent each exploit a different stage of authentication or authorisation. This guide explains how these attacks work, which controls help mitigate them, and why phishing-resistant authentication should be a priority.

4 June 2026
Updated 18 November 2026
9 min read
Key takeaways
  • Stolen and reused credentials remain one of the main ways attackers get in, but their share varies by dataset and year. Verizon's 2026 DBIR reports vulnerability exploitation overtaking credential abuse as the leading initial access vector.
  • AiTM proxies relay your credentials and MFA response to the real sign-in service, capture the session cookie and show the user a normal-looking login. SMS codes, TOTP and push approvals can all be relayed this way.
  • FIDO2 passkeys and security keys resist AiTM phishing because each credential is scoped to the legitimate site's relying party ID and the browser checks the page origin. A lookalike domain cannot obtain a valid response.
  • FIDO2 does not protect a session token stolen after a legitimate login. Against token replay, the main Microsoft controls are Token Protection (device-bound sign-in session tokens, for supported apps and platforms only), CAE for supported apps, endpoint protection, and fast detection and revocation.
  • Number matching, now on by default for Microsoft Authenticator push, reduces accidental approvals in MFA fatigue attacks. It is not a guarantee: an attacker who talks the user through the prompt can still succeed.
  • Prioritise phishing-resistant MFA for privileged accounts, lock down enrolment and recovery so they cannot be used as a back door, restrict OAuth consent, and have a tested response for suspected token theft.

Credentials remain a major breach vector

Breach reporting through the 2010s centred on malware: ransomware groups, nation-state implants, zero-day exploits. Attackers have since put far more effort into credential theft. It is cheap, it scales, and it is hard to spot. A legitimate account signing in through an authorised identity provider looks like a normal login to most monitoring systems.

How large a share credentials account for depends on the dataset. Verizon's DBIR has for years linked stolen credentials to a large majority of hacking-related breaches, which is a narrower category than all breaches. Its 2026 edition reports vulnerability exploitation overtaking credential abuse as the most common initial access vector, while credential misuse still shows up at later stages of many attacks. The fair reading: credentials are one of the main vectors, not the only one.

Supply is not a problem for attackers. Breach aggregators such as Have I Been Pwned index billions of breached accounts. That figure counts account records exposed in breaches, not unique people or credentials that still work, but infostealer logs add fresh passwords and session cookies on top. Stuffing tools test these at scale with throttling and IP rotation to avoid rate limits.

MFA stops most credential stuffing and automated password attacks. Attackers adapted. Some techniques now target the session created after sign-in, or the user's consent, rather than the password. SMS codes, TOTP and push approvals offer little protection against those.

Reading breach statistics

Headline percentages depend on what is counted: all breaches or only hacking-related ones, the first step of an attack or any step, one year or several. When a report quotes a credential figure, check its scope before you put it in a business case.

Six techniques that compromise identities or get around MFA protections

Not all of these defeat MFA directly. Some steal what MFA produces (a session), some exploit a weak MFA method, and some work where MFA is missing or never challenged. A control that stops credential stuffing may do nothing against session token theft.

🕸
1. Adversary-in-the-Middle (AiTM) proxy
High Risk
The attacker sets up a reverse proxy (Evilginx, Modlishka) on a phishing domain. The user enters credentials and completes the MFA prompt as normal. The proxy forwards everything to the real sign-in service in real time, captures the session cookie it returns, and passes the logged-in page back. The user sees nothing wrong. The attacker now holds a valid session and no longer needs the password or the MFA method. Blocked by: phishing-resistant MFA such as FIDO2 passkeys or security keys (the credential is scoped to the real site, so the browser will not produce a valid response for the phishing domain). Mitigated by: Conditional Access policies that require a compliant or Entra-joined device, because the proxied sign-in comes from attacker infrastructure that cannot present that device; Token Protection for supported apps; and alerts on risky sign-ins.
🍪
2. Pass-the-cookie / session token replay
High Risk
Infostealers such as RedLine and Vidar copy session cookies and tokens from browser profiles on an infected device. The attacker imports them into their own browser and carries on with the authenticated session. No password and no MFA prompt, for as long as the session is accepted. Device compliance alone does not stop this: Conditional Access checks the device when a token is issued, but does not bind that token to the device, so a token stolen from a compliant laptop can be replayed elsewhere. Reduced by: Conditional Access Token Protection, which requires device-bound sign-in session tokens for supported apps and platforms; endpoint protection against infostealers; Continuous Access Evaluation (CAE), which lets supported apps act on events such as account disablement, a password reset or a change of network location; and alerts on unusual sign-ins followed by session revocation.
🔔
3. MFA fatigue (push bombing)
Medium Risk
The attacker already has the password and triggers push prompt after push prompt on the target's phone. Uber's 2022 breach began this way: a contractor received repeated prompts and, after a message from someone posing as IT support, accepted one. Mitigated by: number matching (the user must enter the number shown on the sign-in screen, which makes reflexive taps much less likely), showing app and location context in the prompt, and alerting on repeated denied prompts. Number matching is now on by default for Microsoft Authenticator push notifications. It still depends on the user: an attacker who calls the victim and asks for the number can get an approval. Phishing-resistant methods remove the approve-or-deny prompt altogether.
📱
4. SIM swapping
Medium Risk
The attacker calls the target's mobile carrier, impersonates them using social engineering or stolen personal data, and moves the phone number to a SIM they control. Every SMS code then goes to the attacker. This matters most where SMS is the only MFA option. Organisations that already use authenticator apps face limited exposure. Mitigated by: replacing SMS with authenticator apps or FIDO2. If SMS cannot be removed yet, carrier PIN codes or port-out locks reduce the risk.
🤖
5. Credential stuffing at scale
Medium Risk
Automated tools test breached username and password pairs against sign-in pages at scale. Most attempts fail against unique passwords and MFA. The ones that succeed exploit reuse: a forum breach from 2019 can still open a company Microsoft 365 account if the user never changed that password. Strictly, this is not an MFA bypass. It succeeds where MFA is missing, or through legacy protocols that cannot prompt for it. Stopped by: MFA on all accounts, blocking legacy authentication, and leaked credential detection in Microsoft Entra ID Protection, which flags accounts whose credentials appear in known leaks.
🔑
6. OAuth consent phishing
Medium Risk
The attacker registers an OAuth application and sends the target a link. The user signs in on the genuine Microsoft or Google page, completes MFA as normal, and then grants the app the permissions it asks for. Nothing is bypassed: the user authenticates correctly and then authorises the attacker's app, which receives delegated access to mail, files or contacts. Changing the password does not remove that access. The grant stays until an administrator revokes it. Mitigated by: limiting user consent to verified publishers and low-risk permissions (or turning it off), enabling the admin consent workflow so users can request apps for review, regularly reviewing existing user-granted and admin-granted consents, and alerting on new grants with high-privilege permissions.

What each MFA type actually stops

The MFA method you choose determines which attacks still work. SMS codes, TOTP and push notifications stop most credential stuffing. None of them stops a real-time AiTM proxy. Knowing where each method's protection ends tells you what to upgrade first.

MFA Type
Credential Stuffing
AiTM Proxy
Session Replay
SMS OTP
✔ Blocks
✘ Bypassed
✘ Bypassed
TOTP (authenticator app)
✔ Blocks
✘ Bypassed
✘ Bypassed
Push + number matching
✔ Blocks
✘ Bypassed
✘ Bypassed
FIDO2 passkey / security key
✔ Blocks
✔ Blocks
▲ Partial*

* FIDO2 blocks AiTM phishing at sign-in but does not protect session tokens after a legitimate login. Session replay needs separate controls: Token Protection where supported, endpoint protection against infostealers, CAE, and detection with session revocation.

What makes FIDO2 different

SMS codes, TOTP codes and push approvals share one property: they can be relayed. The user completes the challenge and the proxy passes it on. The session opens. The proxy takes the cookie.

FIDO2 works differently. When a passkey or security key is registered, the credential is scoped to the service's relying party ID (for Microsoft Entra, login.microsoft.com), and the private key never leaves the authenticator. At sign-in, the browser checks that the page's origin matches that relying party ID before it lets the authenticator respond. The authenticator then signs the server's challenge together with data bound to the relying party. A phishing page on login-microsoft-365.com cannot ask for a response for login.microsoft.com, so the proxy has nothing valid to relay.

NIST SP 800-63B treats this kind of binding to the genuine service as phishing resistance. Both CISA and the NCSC recommend phishing-resistant MFA for high-value accounts.

Enrolment and recovery decide how strong it really is

Phishing-resistant MFA is only as strong as the way people enrol and recover. If the helpdesk resets MFA after a caller answers a few questions, if Temporary Access Passes are issued without proper identity checks, or if users can still fall back to SMS or push, attackers simply target that path. Require strong identity verification for resets, restrict who can issue Temporary Access Passes and keep them short-lived, and remove weaker methods for accounts that have moved to passkeys.

A note on push MFA

A user who receives repeated push requests may approve one just to stop the noise. Number matching, now enabled by default for Microsoft Authenticator push, makes a reflexive approval much less likely because the user has to type the number shown on the sign-in screen. It does not help if the attacker persuades the user to read out or enter the number. Treat it as a baseline, and move high-risk users to phishing-resistant methods.

How to defend against these attacks

Phishing-resistant sign-in closes the AiTM route, but stolen sessions need controls of their own. Our companion guide on defending against MFA bypass covers session-layer defences such as Token Protection and CAE, the accounts to protect first, seven controls to implement now and what to do if you suspect token theft.

Ryland Deakin
About the author
Lead Consultant, Cyvra · CISM · CompTIA Security+ · MCP

Ryland has delivered cybersecurity, compliance, and IT management programmes for regulated organisations across the UK and the Netherlands for over 20 years. View full profile

Common questions

What is an adversary-in-the-middle attack?

An AiTM attack runs a reverse proxy between the user and the real sign-in page. The user visits a phishing URL, enters credentials and completes the MFA prompt. The proxy forwards all of it to the real site in real time, collects the session cookie, and keeps a valid session with no further need for the user's password or MFA method. Tools such as Evilginx and Modlishka automate this. SMS, TOTP and push MFA all fail here because the authentication completes legitimately through the proxy.

Does FIDO2 stop adversary-in-the-middle attacks?

Yes, at sign-in. FIDO2 passkeys and security keys are scoped to the relying party ID they were registered for, and the browser checks the page origin before the authenticator responds. A passkey registered for login.microsoft.com will not produce a valid response on a lookalike domain. NIST recognises this as phishing resistance; TOTP does not qualify. The protection depends on enrolment and recovery too: weak helpdesk resets or fallback methods give attackers another way in.

What is the difference between credential stuffing and a targeted AiTM attack?

Credential stuffing is automated and untargeted. Attackers run breached username and password lists against common services at scale, exploiting password reuse. Unique passwords and MFA stop most of it. An AiTM attack is targeted and active: the attacker phishes a specific user, runs a proxy to intercept the sign-in in real time, and walks away with a valid session token. SMS, TOTP and push MFA do not stop it; countering it requires phishing-resistant authentication, device-based Conditional Access and session anomaly detection.

Identity Security Assessment

Find out if your MFA can be bypassed

We test your Conditional Access policies, authentication strength configuration, and session controls against current AiTM and credential attack techniques and give you a prioritised fix list.