Moving a company's sign-in to an external identity provider is a security decision. It puts multi-factor authentication, device checks, and offboarding in one place, and it holds only if every other way in is closed. In ZITADEL, a password set before the move could still get a user in.

We found and reported this issue. This post explains what it allowed, how ZITADEL fixed it, what teams running ZITADEL should do, and what I think it teaches anyone who relies on a login policy.

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. Each organization in ZITADEL has a login policy, and one of its settings turns off sign-in with a username and password, so that the organization's users must sign in through an external identity provider instead.

What the issue let an attacker do

In versions 4.0.0 through 4.20.2, that setting was not enforced on every password sign-in. Login V2, ZITADEL's hosted sign-in screen, checked it when a user entered their login name, but its password step did not reliably check it against the user's own organization. The Session API, the application programming interface (API) on which custom sign-in screens are built, accepted password checks without reading the login policy at all.

So an attacker who knew the password of a user in such an organization could:

  • Sign in with that username and password through Login V2, and get a valid session for every application that uses ZITADEL for sign-in.
  • Get around the controls of the identity provider the organization relies on, such as its multi-factor authentication, conditional access, or device checks.
  • Keep signing in after the user was offboarded at the identity provider, as long as the user was still active in ZITADEL and had a password.
  • Sign in through custom sign-in screens built on the Session API that did not check the setting themselves.

The issue did not bypass the password itself. The password was still verified, and failed attempts still counted toward the lockout policy. What it defeated was the identity provider's controls: a known or leaked password was enough. It affected organizations that had turned off password sign-in while some of their users still had a password in ZITADEL, for example one set before the organization moved to an external identity provider. Users without a password were not affected, and neither was Login V1. ZITADEL rated the issue high severity, 7.1 out of 10 on the Common Vulnerability Scoring System (CVSS).

A fix that covered one path

The advisory describes this issue as an incomplete fix of an earlier one, GHSA-25rw-g6ff-fmg8 (CVE-2026-29193). That fix enforced the setting on the Login V2 server but did not cover every path that accepts a password.

How it was reported and fixed

We reported the issue to ZITADEL through coordinated disclosure, and the advisory credits the report to Robert Capps of Stratovera. ZITADEL fixed it in version 4.20.3 and published advisory GHSA-gx82-3gqq-j4h5 on November 13, 2026.

ZITADEL now checks the login policy of the user's own organization on every password check through the Session API, and Login V2 checks the same setting at its password step. If the user's organization or its login policy cannot be determined, the password check is refused.

What teams running ZITADEL should do

  • Upgrade to version 4.20.3 or later. The advisory names no configuration that fully closes the gap on older versions.
  • Update custom sign-in screens. After the upgrade, password checks for users whose organization requires an external identity provider fail. The advisory asks screens built on the Session API to send those users to their identity provider.
  • Find the passwords left behind. The users this issue exposed are those who still had a password in an organization that no longer allows one. I would find out who they are.

What it teaches about login policies

These are my conclusions, not the advisory's.

A policy is enforced where credentials are checked, not where it is configured. Turning a setting off in an admin screen is a statement of intent. It protects you only if every path that accepts that credential asks the same question, including the programming interfaces that custom screens use.

When one path is fixed, ask which others accept the same credential. This issue followed a fix that covered one server and not every password path. A security review of a fix should list every way in, not only the one in the report.

Retiring a kind of credential is not finished until the old credentials are gone or refused everywhere. Moving sign-in to an identity provider changes where new users sign in. It does not remove the passwords that existed before the move.

If your company relies on an identity platform's policies, we can review whether they hold on every path. Talk to us about a security review, or see all of our security advisories.