跳到主要内容

Origin binding

Origin binding is the single most important thing ZTXBAS does. It's why a phished user cannot approve an attacker's login.

The problem it solves​

Traditional push-based MFA looks like this:

  1. User types their password on paypal-secure.example.com (a phishing site).
  2. Attacker forwards the credentials to the real paypal.com.
  3. Real PayPal sends a push to the user's phone: "Approve login?"
  4. User taps Approve because they just tried to log in.
  5. Attacker is now inside.

The phone doesn't know which site the login attempt came from. The approval prompt has no origin to display, so the user has nothing to compare against.

What ZTXBAS does​

Every challenge is bound to an origin at creation time, and the origin is what the user sees on their phone.

ZTXBAS

The RP can only create a challenge for an origin that was previously registered via POST /v1/origins. A challenge for https://phishing.example.com fails with 403 UNREGISTERED_ORIGIN because no honest operator ever registered that origin.

The three checks, in order​

When you call POST /v1/auth/challenge, ZTXBAS runs three checks:

  1. HMAC verifies. The request is signed by an application secret. A phishing site doesn't have your secret.
  2. The origin is registered. The RP-declared origin exists in origins for this application. A phishing site couldn't have registered its own origin under your application because it has no admin access.
  3. The user exists. Prevents challenge storms against enumeration.

Only then does the mobile push fire.

Origin normalization​

Origins are normalized before comparison:

  • Scheme and host are lowercased.
  • Default ports are stripped (:443 for https, :80 for http).
  • Trailing paths are rejected - an origin is scheme://host[:port], never with a path.

This means HTTPS://APP.example.com:443 and https://app.example.com are the same origin. The canonical form is what gets bound into the challenge and echoed back in the JWT's origin claim.

The JWT's origin claim​

After approval, the JWT carries the canonical origin in its origin claim. Your backend can (and should) cross-check that this matches the origin you expected:

if claims.Origin != "https://app.example.com" {
    return errors.New("token issued for a different origin")
}

This closes a subtle attack where an attacker replays a JWT they obtained legitimately (for their own account) at a different origin they control.