Multi-factor authentication

Passwords get phished, reused across sites, and guessed. Multi-factor authentication (MFA) asks for a second, independent proof, so a stolen password or a compromised mailbox alone is not enough to take over an account. Better IAM supports authenticator apps (TOTP) with recovery codes, passkeys, and, where an organization allows it, one-time codes sent to a verified email address. A browser that completed MFA can be remembered for a while so people are not asked on every sign-in.

When MFA is required

You rarely switch MFA on per person; rules make it required. MFA is required when any of these holds:

• the person is a root administrator (always, and never through a remembered device or an emailed code);
• the person has an authenticator enrolled;
• the tenant policy sets requireMfa, or requireMfaForOwners and the person is an owner;
• the deployment's authentication.requireMfa(tenant, identity) callback returns true.

Use the tenant policy for rules an organization decides for itself. Use the callback for rules that depend on your own logic, such as a plan tier or a tenant type:

authentication: {
  // Adds to (never replaces) each tenant's own requirement.
  requireMfa: (tenant, identity) => tenant.type === 'organization' && identity.owner,
},

The requirement applies to password, passwordless, and federated sign-in alike, and to sessions that already exist: they are re-validated on every use, so requiring MFA in a tenant locks out sessions that did not complete it (MFA_REQUIRED) on their next request. A person cannot disable MFA while their tenant or the root role requires it. Resetting a password does not remove MFA.

Completing a challenge

After a correct password (or magic link, or federated sign-in), a person who needs MFA is not signed in yet. The first factor returns a challenge, a short-lived ticket that proves the first step succeeded, instead of a session. Your UI then asks for the second factor and sends it with the challenge:

type MfaRequired = {
  mfaRequired: true;
  challenge: string;
  enrollmentRequired: boolean; // no authenticator enrolled yet
  emailCodeAvailable?: boolean; // requestMfaCode may email a code
  passkeyAvailable?: boolean; // a registered passkey may answer
};

The challenge is single-use and valid for five minutes. What the person does next depends on the flags:

const session = await client.auth.verifyMfa({
  tenantId,
  challenge: result.challenge,
  code: '123456',
  rememberDevice: true, // optional; see below
});

A wrong authenticator code, emailed code, or recovery code for a real account is recorded as a failed attempt (auth:signin:fail with reason: 'mfa' or 'recovery-code'). Second-factor attempts are rate limited on the sensitive tier both per challenge and per person, so a correct password does not buy a fresh budget of code guesses.

Authenticator apps (TOTP)

An authenticator app on a phone or in a password manager generates a six-digit code that changes every 30 seconds (TOTP, time-based one-time passwords). It works offline, costs nothing to send, and is the factor most people already know.

• auth.beginMfa generates a secret and an otpauth:// URI labeled with the person's email and your appName. The secret is stored with authenticated encryption.
• Enrollment must be confirmed within ten minutes, with the same credential that started it: the sign-in challenge, or a recently authenticated session.
• Codes are six digits on 30-second steps, and a code from the adjacent step is accepted to absorb clock drift. Each time step is accepted once, so a captured code cannot be replayed.
• A person who already has an authenticator gets MFA_ALREADY_ENABLED from beginMfa, and cannot start a new enrollment from a sign-in challenge (MFA_REQUIRED): they must use their existing factor first.

People who are not required to use MFA can still turn it on from their account page. They enroll from a signed-in session, which needs recent authentication:

const { uri } = await client.auth.beginMfa(); // no challenge: the session is the credential
const { recoveryCodes } = await client.auth.confirmMfa({ code: '123456' });

Confirming an enrollment ends every existing session of the person and returns a new, MFA-verified one. Through the HTTP handler the new session replaces the cookie.

Recovery codes

Phones get lost and reset. Recovery codes are the way back in without an administrator: enrollment returns ten single-use codes, which the person should store somewhere safe, such as a password manager. They are stored hashed and consumed on use.

• auth.recoverMfa({ tenantId, challenge, code }) completes a sign-in challenge with one (audited as auth:mfa:recover).
• auth.regenerateRecoveryCodes() replaces the whole set. It needs recent authentication and an MFA-verified session (audited as auth:mfa:recovery-codes).
• auth.mfaStatus() reports recoveryCodesRemaining, so your account page can prompt people to regenerate before they run out.

Emailed one-time codes

Emailed codes let an organization require MFA without forcing every member to install an authenticator. They are off by default; turn them on per tenant with mfaEmailCodes or for the deployment with authentication.mfaEmailCodes (the tenant setting wins).

• A sign-in offers them (emailCodeAvailable: true) only to people with nothing enrolled and a verified email address, only when sendEmail is configured, and never to root administrators.
• auth.requestMfaCode emails a six-digit code (template mfa-code) bound to that challenge. It is hashed at rest, single-use, and valid for at most ten minutes and never beyond the challenge itself. Requesting again replaces the code, and requests are rate limited per challenge and per person so repeated sign-ins cannot flood a mailbox.
• A challenge that did not offer codes fails with FEATURE_DISABLED.

Passkeys as a second factor

Whenever a person has a passkey registered, the challenge carries passkeyAvailable: true and they can complete MFA with it instead of typing a code. The assertion is verified like a passkey sign-in and bound to the pending sign-in challenge, which must still be open and belong to the same person; both challenges are consumed together. A passkey sign-in on its own already satisfies MFA. See passkeys.

Remember this device

Asking for a code every time someone signs in from the same laptop is tedious, and tedium pushes organizations to turn MFA off. "Remember this device" lets a browser that recently completed MFA skip the second factor for a while, while a new device, or a stolen password used elsewhere, still has to pass it.

rememberDevice: true on verifyMfa, confirmMfa, or finishPasskeyMfa asks Better IAM to remember the browser. When the deployment and the tenant allow it, the response carries a deviceToken and deviceExpiresAt. Passing that token to signIn or finishPasswordless later satisfies the MFA requirement without a code.

• Lifetime. The shorter of authentication.trustedDeviceLifetimeMs (30 days by default, one year at most, 0 disables the feature) and the tenant's trustedDeviceDays (0 turns it off for the tenant).
• Storage. The token is opaque and stored hashed. Through the HTTP handler it lives in its own HttpOnly better-iam.device cookie (__Host- prefixed over HTTPS), which the handler injects into later signIn and finishPasswordless bodies, so the client only has to send rememberDevice: true once.
• Root administrators are never remembered.
• Invalid or expired tokens simply lead to the normal challenge.
• Sessions established this way carry trustedDeviceId. They count as MFA-verified (principal.mfa) but carry no fresh factor time (principal.mfaTime is absent).

People manage their remembered devices from their account page:

• auth.listTrustedDevices lists the live ones, newest use first, with the client details, createdAt, lastUsedAt, and expiresAt, so people can recognize each browser.
• auth.revokeTrustedDevice({ deviceId }) forgets one device, for example a shared computer. The next sign-in from it asks for MFA again. Needs recent authentication.
• auth.revokeTrustedDevices() forgets all of them, and through the HTTP handler also clears the device cookie. Needs recent authentication.

const devices = await client.auth.listTrustedDevices();
await client.auth.revokeTrustedDevice({ deviceId: devices[0].id });

Every remembered device is forgotten when the person changes or resets their password, changes their email, enrolls or disables an authenticator, or removes a passkey, when an administrator revokes their sessions or everyone's in the tenant, and on revokeTrustedDevices. Remembering and forgetting are audited as auth:device:trust and auth:device:revoke.

Managing factors

An account page usually shows what a person has set up and lets them change it. These calls work from a signed-in session:

Step-up

Some actions deserve more than a valid session: refunding a payment, deleting a project, changing who is an owner. Step-up authentication asks the person to prove themselves again, or with MFA, right before such an action. Better IAM gives you three building blocks:

• Recent authentication. Sensitive operations require a session established within recentAuthenticationMs (five minutes by default) and fail with RECENT_AUTH_REQUIRED otherwise. auth.reauthenticate({ password }) re-runs the password ceremony, and the second factor when the account has one, and issues a fresh session. See recent authentication.

• Policy conditions. Policies can read facts about the principal (the caller): principal.mfa (the session passed MFA), principal.mfaTime (when a first-hand factor was last verified; absent for remembered-device, impersonation, and API-key sessions), and principal.authMethod. A policy can demand MFA for one action only:

  {
    effect: 'allow',
    actions: ['billing:refund'],
    resources: ['invoice/*'],
    conditions: { Bool: { 'principal.mfa': true } },
  }

• Framework guards. The framework integrations accept a stepUp requirement (mfa: true, mfa: 'fresh', or maxAgeMs) on pages, routes, and actions, and redirect to your stepUpPath with a reason. See Next.js guards.

Just-in-time elevation can also require an MFA-verified session for every activation of an administrator role; see elevation.