Multi-factor authentication options compared: what each one protects against
Compare MFA methods, phishing resistance, recovery risks and factor-change controls using OWASP's published guidance.
OWASP's Multifactor Authentication Cheat Sheet explains what MFA is, when to require it and the strengths and weaknesses of each method. We read it on 15 November 2026 through its "something you do" section, and not the last sections or the NIST, passkey and other sheets it links. Everything attributed below is OWASP's, which in turn cites NIST SP 800-63B-4 in places.
What counts as MFA
Evidence from at least two distinct factors: something you know, something you have, something you are. Two things of the same factor, such as a password and a PIN, are not MFA and add little. Location and other risk signals can inform decisions but are not a factor. OWASP adds that many weaknesses it lists matter mainly against targeted attacks, and that any MFA is better than none.
Why bother
OWASP says the most common way accounts are compromised is weak, reused or stolen passwords, and that designers should assume passwords will be compromised. It calls MFA the best defence against most password attacks (brute force, credential stuffing, password spraying) and cites a Microsoft analysis; we do not repeat its figure here because we did not read that analysis.
Costs it lists: more complexity for admins and users, hardware cost, lockouts when a factor is lost, external dependencies that can fail or be vulnerable, and reset processes that attackers can abuse.
Quick recommendations
Require MFA for all users; offer TOTP; require it for admins and other privileged users; build a secure way to reset MFA; consider MFA as a service. How and when depends on the app's threat model, users and control over users.
When to ask for it
At login, and for sensitive actions: changing the password or security questions, changing the account email, disabling MFA, or elevating to an admin session. If there are several ways to sign in (a separate login API, a mobile app) each needs MFA or other protection; OWASP says these are often missed.
The methods
| Method | OWASP's main points |
|---|---|
| Password or PIN | Simple, but weak and reused passwords, and phishable. |
| Security questions | OWASP says NIST no longer recognises them as an acceptable factor. Answers are guessable or found on social media. |
| Hardware OTP token | Hard to compromise remotely; costly and slow to replace; a stolen one works without a PIN; phishable though short-lived. |
| Software OTP (TOTP app) | Cheap, standard, new app can be set up without shipping hardware. RFC 6238's default step is 30 seconds. Phone can be lost, compromised or be the same device used to sign in; backup seeds may be stored insecurely; phishable. |
| U2F / security keys | Phishing-resistant through relying-party/origin binding when the protocol is verified correctly; cost, and easy to forget. |
| Client certificates | Phishing-resistant, centrally revocable; need PKI, hard to install, can be stolen from a compromised machine. |
| Smart cards | Unusable without the PIN; need PKI and third-party software in browsers. |
| SMS and phone calls | NIST designates them restricted; OWASP says not to use for high-value or PII-handling apps. Risks: SIM swap, interception, preview on lock screen, message-cost abuse. If it is the only option, OWASP says to document the risk, rate limit per account, watch for SIM-swap signals and plan to move to TOTP, push or WebAuthn/FIDO2. |
| Email codes | Easy, but only as strong as the email account, which often lacks MFA. OWASP notes debate over whether it is MFA. |
| Biometrics | Can be convenient, but spoof resistance depends on the implementation; they need hardware, raise privacy concerns, and cannot be changed if compromised. |
| Location (IP, geolocation, geofence) | A risk signal, not a factor. Bypassable with IPs in the trusted country; VPNs and relays reduce accuracy. |
| Behavioural signals | Risk input, not a standalone factor. |
Passkeys
OWASP says passkeys can give phishing-resistant MFA, but when used as MFA you must require user verification and check the verification flag on the server; a touch showing only presence is not a second factor. Its separate Passkey Security Cheat Sheet, which we did not read, covers the details.
One-time codes
For generated delivery codes, use short lifetime, single use, strict attempt limits, invalidate after success, no logging, no long-term plaintext. OWASP also recommends a secure random generator, considering 8 digits or more where usable, and replacing the old code on resend. It says hashing OTPs adds little against an attacker with the database (small keyspace), but still prevents accidental exposure through logs and limits the damage from brief database exposure.
TOTP is different from a generated delivery code: it requires a protected long-lived shared seed, which cannot simply be replaced by a hash. Store that seed encrypted with restricted key access; enforce replay and attempt controls.
Making it livable
- Risk-based prompts (new device or location) reduce how often users are asked, since constant prompts can lead users to disable MFA.
- If the second factor fails, offer another method or a reset, and notify the user of the failed attempt with time, browser and location.
- Resetting MFA is the hard part: recovery must not become a bypass. OWASP's options: single-use recovery codes at setup, several MFA types, a one-use code by post, a support process with rigorous identity checks, or another trusted user vouching. It says there is no single best way.
- Changing a factor (new phone, new authenticator): require re-authentication with an existing factor, do not trust the session alone, treat it as high risk, notify out of band, consider delays for high-value accounts.
- MFA as a service: useful if you lack resources, but a compromise of the service could bypass MFA for every app using it.
Questions for your team
- Which of our users and actions require MFA, and does every sign-in route enforce it?
- Which method do we offer first, and why?
- How does a user who lost their phone get back in, and could an attacker use the same route?
- Do we send a notification when a factor changes?
Related: forgot-password flows, session management basics, app authentication options.
Limits
One OWASP page, partly read. No pricing, adoption or effectiveness figures are given by us. Regulatory duties for MFA (the page mentions finance, healthcare and GDPR) vary and need counsel for your case.
By Omnistu.