A stolen password alone no longer opens an account protected by multi-factor authentication, because the password satisfies only the first of at least two independent checks and the second demands something the thief didn't take. The practical damage shrinks to what the attacker can do inside one authenticated session before that second factor expires or the user notices. Password-only breaches usually mean full, repeatable access; MFA breaches usually mean a narrow window, a single device, or a single service β and often nothing at all.
That distinction matters more than most security advice admits. MFA doesn't make a stolen password worthless. It makes it partial. The attacker holds half a key, and the lock still needs the other half β unless the attacker can trick a human into turning it, which is the failure mode this article walks through in detail.
The Two-Lock Model: What a Password Proves and What MFA Adds
A password proves one thing: that whoever typed it knows a string of characters. Multi-factor authentication adds a second proof from a different category β something you have, like a phone app or hardware key, or something you are, like a fingerprint. The categories are deliberately separate so that stealing one doesn't hand over the other.
When an attacker buys a password from an infostealer log or a credential-stuffing run, they hold the "knows" factor. The "has" factor lives on hardware they don't control. That's the entire mechanism in one sentence: the attacker has one of two required proofs, and the system won't proceed until it sees both.
The proof doesn't need to be exotic. A one-time code from an authenticator app counts because the server accepts it only once and only within a short validity period β commonly 30 seconds for time-based codes under the TOTP standard maintained by the Internet Engineering Task Force. A push notification counts because it lands on a registered device. A passkey counts because the private key never leaves the user's hardware.
What a password alone grants is replayability. The attacker can use it today, tomorrow, and from any IP address until someone changes it. MFA removes replayability from the equation. The stolen string becomes a ticket that's only valid at the gate, in person, with ID β and the thief doesn't have the ID.
Why a Stolen Password Stops at the Second Gate
Think about what happens at login. The attacker submits the username and password. The server checks them against its stored hash and finds a match. In a password-only world, that's the end of the conversation β the server issues a session token and the attacker is in. With MFA enabled, the server pauses and issues a challenge instead. The attacker now has to satisfy a requirement they didn't prepare for.
Most credential theft happens at scale and without the victim present. Infostealer malware grabs saved browser passwords and session cookies from thousands of machines at once. Credential-stuffing botnets spray username-password pairs harvested from unrelated breaches across hundreds of services. Neither method gives the operator a live person holding a registered phone. So the challenge simply doesn't get answered.
The attacker's options when the challenge appears are narrow. They can abandon the attempt, which is what happens in the overwhelming majority of cases. They can try to guess a code, which fails because codes are short-lived and rate-limited. Or they can pivot to social engineering β calling the victim, impersonating IT, asking them to read back a code β which is a different attack requiring a human on the other end and a story that survives scrutiny.
That pivot is where breaches actually happen, and it's worth naming the pattern plainly: real-world MFA bypasses almost never defeat the cryptography. They defeat the human, the help desk, or the enrollment flow. The second gate holds unless someone opens it from the inside.
The Numbers That Show the Gap: Microsoft's 99.9 Percent Figure
Microsoft has published, repeatedly in its security blog and annual Digital Defense Report, that accounts with MFA enabled are more than 99.9 percent less likely to be compromised than accounts without it. That figure has been cited since at least 2019 and remains the most quoted single data point in the MFA conversation. Treat it as directional, not literally predictive β it compares enabled versus disabled populations, not identical users with one variable flipped.
Read the percentage carefully and it says something specific: the compromise rate for MFA-protected accounts in Microsoft's telemetry is roughly one-thousandth of the rate for unprotected accounts. That's not a guarantee any individual account is safe. It's a statement about populations. Somebody still has to be the 0.1 percent, and the attack paths that produce those cases are well documented.
A second figure worth holding in mind comes from Verizon's annual Data Breach Investigations Report. Across recent editions, the "human element" β phishing, pretexting, misuse β appears in roughly two-thirds to three-quarters of breaches. Stolen credentials are the most common entry vector. Both facts point the same direction: passwords are the commodity attackers trade in, and humans are the door they knock on when the technical path closes.
Put those two statistics side by side and the arithmetic is sobering. If credentials are the main way in, and MFA cuts compromise rates by three orders of magnitude for accounts that use it, then the marginal value of turning MFA on for a password-exposed account isn't a modest improvement. It's the difference between a commodity breach and a targeted one.
Why Reused Passwords Still Fail Under MFA β and Where Reuse Bites
Here's the counterintuitive part. MFA doesn't rescue password reuse. If you use the same password on twenty sites and one of them leaks, the attacker holds a key to nineteen doors, and MFA only blocks the ones where it's actually enabled. The unprotected services fall first, and what falls out of them often feeds the attack on the protected ones.
Picture the sequence. A forum you signed up for in 2014 leaks its user table. The attacker takes your email and password and tries them on your bank, your email provider, and your employer's VPN. Your bank has MFA β blocked. Your email provider has MFA β blocked. The VPN has MFA β blocked. But the defunct shopping site you also used that password on has no MFA, and inside it sits a password-reset email trail pointing at your primary address, plus a support ticket with your mother's maiden name.
Password reuse converts a single leak into a reconnaissance map. MFA protects the gates it's installed on, but the surrounding terrain β recovery email, security questions, linked accounts, saved payment methods β is often undefended. A competent attacker doesn't bang on the MFA-protected door. They walk around it.
The practical rule is blunt. Enable MFA everywhere it's offered, starting with email, banking, and any account that can reset the others. Then stop reusing passwords, because reuse creates the side doors MFA can't see. A password manager generating unique strings per site closes that gap; the built-in managers in browsers and phones are adequate for most people, and dedicated tools add sharing and audit features.
Five Bypass Routes That Get Past the Second Gate
None of these defeat MFA cryptographically. Every one goes around it, and knowing the routes makes the defenses obvious.
1: Real-time phishing proxies. Tools like Evilginx sit between the victim and the real login page, relaying the password and the MFA code in real time, then stealing the resulting session cookie. The attacker never needs the second factor β they need the victim to supply it once, to a page that looks right. 2: MFA fatigue (push bombing). The attacker has the password and spams push notifications until the exhausted user taps Approve. Uber's 2022 breach followed this path, reportedly after the attacker obtained credentials from a contractor's machine and then bombarded them with pushes. 3: Help-desk social engineering. The attacker calls support, claims to have lost their phone, and talks a human into re-enrolling a new device. MFA is bypassed by resetting it β no code needed. 4: SIM swapping. If the second factor is an SMS code, the attacker convinces the carrier to port the victim's number to a SIM they control. The FCC has documented this fraud extensively and pushed carriers toward stronger authentication for port requests. 5: Session-cookie theft. Infostealers grab the cookie that proves an already-completed login. The attacker imports it and skips the login flow entirely, MFA included, until the session expires.
Each route requires something a bare password doesn't: a live victim, a helpful employee, a compliant carrier, or malware on an authenticated machine. That's the point. The stolen password alone is no longer enough; the attacker has to acquire a second capability on top of it.
Push, TOTP, SMS, or Passkey: What Each Survives
Not all second factors resist the bypass routes equally. The table below compares them on the dimensions that decide real-world outcomes: what an attacker needs to defeat each, whether phishing can capture it, and what recovery looks like when a device is lost.
| Factor type | Defeats a stolen password alone? | Defeated by real-time phishing? | Defeated by SIM swap? | Typical recovery path |
|---|---|---|---|---|
| SMS code | Yes | Yes | Yes | Carrier port freeze, then re-enrollment |
| TOTP app code | Yes | Yes (via proxy relay) | No | Backup codes or account recovery |
| Push notification | Yes | Partially (needs victim tap) | No | Backup device or recovery key |
| Passkey / FIDO2 key | Yes | No (origin-bound) | No | Second device or recovery flow |
The pattern is clear. SMS is the weakest because the second factor lives on a channel the attacker can hijack through the carrier. TOTP resists SIM swaps but falls to a proxy that relays the code in real time. Push resists both but depends on the user not tapping Approve out of habit. Passkeys, built on the FIDO2 standard from the FIDO Alliance, bind the credential to the site's origin, so a proxy on a lookalike domain can't complete the ceremony.
The verdict is conditional. If you can only do one thing, enable any MFA β a weaker factor beats none. If you can choose, move from SMS to an authenticator app, then to push with number matching (which shows a number on the login screen the user must type into the app, defeating blind push fatigue), then to passkeys or hardware keys for email, banking, and administrative accounts. Number matching is a meaningful improvement over plain push and is now the default in Microsoft Entra ID.
Attacker Economics: What a Password Is Worth With and Without MFA
Stolen credentials are priced on criminal marketplaces by how much access they buy. A verified password-only account for a consumer service typically trades for a small fraction of the value of one that also carries a session cookie, a linked payment method, or administrative access. Automated pricing lists on these forums move constantly, but the tiering is stable: plain credentials at the bottom, "full logs" from infostealers β cookies, saved passwords, autofill data β priced several times higher.
The MFA premium is the reason. An account with MFA enabled is harder to monetize from a password alone. A buyer who spends money on a credential that dies at the second gate has wasted it. So sellers either strip such listings, bundle them with session cookies that skip the gate, or sell them cheap to buyers who intend to run a social-engineering play. MFA changes the market structure of the stolen credential, not just its price.
There's a second-order effect worth noting. When MFA is widespread, credential-stuffing botnets produce fewer usable hits per thousand attempts, so attackers shift toward phishing and infostealers that capture post-MFA material. That's the arms race described in Verizon's report and in Microsoft's threat intelligence writing: MFA doesn't eliminate credential theft, it raises the cost and pushes attackers toward the human. For defenders, that means MFA without phishing-resistant factors is a speed bump, not a wall.
Where MFA Fails Completely: Recovery Flows and Legacy Protocols
Every MFA deployment has a back door called account recovery. If a user can reset their password by answering security questions, receiving an email to a non-MFA mailbox, or calling a help desk, the attacker can skip MFA entirely by going through recovery. Help-desk reset fraud is the single most common way MFA gets defeated in enterprise breaches, and it's largely a process problem, not a technology one.
Legacy protocols make it worse. Older email authentication methods like IMAP, POP, and SMTP basic auth often can't handle an MFA challenge, so organizations leave app passwords or "legacy auth" enabled to keep old clients working. Those app passwords bypass MFA by design. Microsoft and Google have both spent years disabling basic auth for this reason; Microsoft's guidance and Google's Workspace defaults now block it by default. When legacy auth stays on, a stolen password plus an app password is MFA-equivalent access.
Service accounts and API tokens carry the same risk. A token issued once and stored in a script doesn't ask for a second factor. If the token leaks, the attacker has standing access until it's rotated. The mitigation is short expiry, scope limits, and rotation on any suspicion of exposure β the same controls that make cloud access auditable under frameworks like the CIS Benchmarks.
The 30-Second Rule: Session Lifetime as the Real Damage Boundary
The duration of a session determines how long a stolen cookie remains useful. Many services issue session tokens that persist for days or weeks until explicit logout, so a cookie stolen today can carry an attacker for as long as the token lasts. Shortening idle timeout and forcing re-authentication for sensitive actions β the pattern banks use for transfers β limits the blast radius.
This is where the arithmetic of a breach gets concrete. A stolen password in a password-only system gives access until the password is changed, which is usually when the victim notices β often months later. A stolen session cookie gives access until the token expires or is revoked, which can be minutes to weeks depending on the service. Short session lifetimes reduce the value of the stolen cookie to the attacker in direct proportion to their length.
Practical guidance: check the session settings on your most important accounts. Many providers allow you to view and revoke active sessions from a security dashboard. If a session shows an unfamiliar location or device, revoke it and change the password. The attacker's window closes the moment you do β and that's the operational meaning of MFA as a damage-limiter rather than a magic shield.
FAQ
Can MFA be bypassed entirely, and how often does it happen?
Yes, and the routes are well documented: real-time phishing proxies, push fatigue, help-desk reset fraud, SIM swapping, and session-cookie theft. None of these break the cryptography β they exploit the human or the recovery process. The frequency varies widely by industry, with high-value targets like executives and IT administrators attacked most often through these paths. The consistent finding across breach reports is that MFA bypass requires an interactive step the attacker must engineer, not just a password they bought.
Is SMS-based MFA still better than no MFA at all?
Yes, by a wide margin, despite being the weakest form. SMS codes defeat credential stuffing and infostealer password dumps completely because the attacker doesn't have the phone. The known attacks β SIM swapping and interception β require targeted effort against a specific victim, which is a different threat model from mass credential theft. The recommendation is to use SMS as a floor, then upgrade to an authenticator app or passkey when the account supports it.
What happens to a stolen password if the account has a passkey instead of a code?
The password becomes nearly useless on its own. A passkey is bound to the site's origin and the user's device, so the attacker can't complete authentication with just the password β the server requires a signed challenge from a key held in hardware. Phishing proxies that relay codes don't work against passkeys because the browser refuses to sign for a lookalike domain. The remaining risks are account recovery flows and fallback passwords if the service still allows them.
Does MFA protect against credential stuffing on unrelated sites?
No. MFA protects only the services where it's enabled. If the attacker takes your leaked password to a site that has no MFA, the second gate doesn't exist. That's why password reuse is the multiplier on any credential breach and why MFA coverage is uneven β the weakest site you've reused a password on is your real security level. Unique generated passwords plus MFA on the important accounts closes both gaps.
Is it worth paying for a hardware security key, or is a phone app enough?
A hardware key is worth it for the small set of accounts that can reset everything else: email, the password manager, banking, and admin consoles. It's phishing-resistant in a way that app codes aren't, because there's no code to relay. For most people, app-based TOTP and number-matched push cover the practical risk. The stronger trigger for buying a key is being a high-value impersonation target, not being an average user.
What should I do first if I suspect my password was stolen but MFA is on?
Check the active sessions list on the affected account and revoke anything unfamiliar, then change the password from a device you trust. Review recovery email and phone numbers for changes the attacker might have made before you noticed. If a session cookie may have been stolen by malware, run a scan before re-authenticating and consider that any passwords saved in the browser should be rotated. Report the incident to the service if there's any sign of account takeover.