The most exposed account in an identity platform is often the one nobody has used yet. An invited user who has not set a password has nothing to steal, so it is easy to assume there is nothing to attack. In ZITADEL, that assumption did not hold: in versions 4.0.0 through 4.19.4, an attacker could claim such an account before its owner did.

Stratovera was one of the independent reporters of the issue. This post explains what it allowed, how ZITADEL fixed it, what teams running ZITADEL should do, and what I think it teaches anyone choosing or running an identity platform.

What ZITADEL is

ZITADEL is an open-source identity platform. Applications use it for sign-in: they send a user to ZITADEL and trust its answer about who that user is. Login V2 is ZITADEL's hosted sign-in screen, where people sign in and where new users set up their first password or passkey.

What the issue let an attacker do

Before a new user sets up a first credential, Login V2 asks them to prove that they control their email address or hold their invite code. It then remembered that proof in a cookie in the browser. As the advisory explains, the cookie's value was built only from values that the browser controls or can learn, with no secret held by the server. An attacker could therefore compute a valid proof for any user without ever seeing the email or the invite code.

With that forged proof, an attacker who knew the user ID or login name of a user with no way to sign in yet could do any of three things:

  • Set the user's password and sign in as them.
  • Register a passkey for the user and sign in with it.
  • Link the attacker's own account at an external identity provider to the user and sign in through it, where an identity provider that allows account linking was configured.

Each path ends the same way: the attacker is signed in as that user, with the user's permissions in ZITADEL and access to every application that uses ZITADEL for sign-in. Invited users, and users created through ZITADEL's application programming interface (API) before their first sign-in, were the ones exposed.

The limits matter as much as the impact. Users who already had a password, a passkey, or a linked identity provider were not affected. Deployments that do not serve Login V2 were not affected, and neither were the ZITADEL APIs themselves. ZITADEL rated the issue high severity, 7.4 out of 10 on the Common Vulnerability Scoring System (CVSS).

How it was reported and fixed

We found and reported the issue to ZITADEL independently, through coordinated disclosure. The advisory credits four independent reporters: Adam Korczynski of Ada Logics, Lucas Dodgson of InfoGuard Labs, CodeAnt AI, and Robert Capps of Stratovera. ZITADEL fixed it in version 4.19.5 and published advisory GHSA-84fg-x5h4-w8xv on October 15, 2026. It is also recorded as CVE-2026-119128.

The fix makes the proof something the server issues itself. Login V2 now accepts a verification proof only after the backend has accepted the email or invite code, signs it with a secret held by the server, binds it to the user and the browser, and lets it expire after five minutes. Proofs in the old format are rejected.

What teams running ZITADEL should do

  • Upgrade to version 4.19.5 or later. The advisory names no configuration that fully closes the gap on older versions.
  • Set the secret that signs the new proof. The advisory asks for a randomly generated value of at least 32 characters, not one copied from an example configuration.
  • Until you can upgrade, leave no active user without a credential. The advisory's partial mitigation is to create users with an initial credential.

What it teaches about identity platforms

These are my conclusions, not the advisory's.

Onboarding is part of the security boundary. Teams test sign-in carefully and treat invitations and first-time setup as product details. Here, the weakness sat in exactly that path, and it reached only users who had never signed in. When you review an identity system, review how an account comes into existence, not only how it is used.

A proof is only as strong as what the attacker cannot know. A value built only from things the client controls is a label, not a proof. For every token, cookie, and link in a sign-in flow, ask what secret makes it hard to forge and who holds that secret. If the answer is "nobody", it is not a proof.

Look at how a platform handles its own flaws. An open-source identity platform's code can be read by anyone, and ZITADEL's advisory is clear about what was affected, what was not, and what to do. When you choose a platform, that record is worth as much as its feature list.

If your company runs its own sign-in, or is choosing an identity platform, we can review how it handles onboarding, sign-in, and recovery. Talk to us about a security review, or see all of our security advisories.