ZITADEL’s hosted Login V2 UI allowed an unauthenticated attacker who knows only a victim’s login name to redirect that victim’s genuine password-reset, invite, and email-verification links to an attacker-controlled host, while the links still carried the real single-use codes. Harvesting such a code lets the attacker set a new password — or complete verification — and sign in as the victim, including an instance administrator.
The login UI built these email links from client-supplied proxy request headers ( x-zitadel-public-host , x-zitadel-forward-host , and x-forwarded-host ) without confirming the host belonged to the instance, and the API’s trusted-domain control — which is meant to reject an untrusted public host — was silently skipped when the submitted host was malformed in a way that resolved to an empty host (for example a value with a non-numeric port such as evil.com:443/ ). Any of these headers can be used; the issue is not specific to one of them. The shared missing invariant is that a public host taken from a request header is never validated against the instance’s trusted domains before it is used to build a security-sensitive link.
An attacker who can reach the unauthenticated Login V2 reset or verification flow and knows a victim’s login name can:
Capture a victim’s single-use password-reset code by poisoning the host of the genuine reset email, then set a new password and take over the account .
Capture invite and email-verification codes through the same primitive, taking over accounts during onboarding or email-change flows.
Take over an instance administrator (for example IAM_OWNER ), resulting in full control of the instance.
Retain access across a later password rotation , because once signed in as the victim the attacker can enrol their own credential or second factor.
This affects applications that authenticate users through ZITADEL’s hosted Login V2 UI (OIDC/SAML); the x-zitadel-public-host and x-zitadel-forward-host headers in particular are not normalized by a standard fronting proxy, so the host can be set by a client even in a typical deployment. The email itself is genuine — correct sender, correct branding, a real code — so only the link host differs; the poisoned link must be fetched once, by the victim opening the link or by automated inbound-mail link scanning.
The attacker does not need the victim’s password, mailbox access, an existing session or any secret, nor any API credentials, localhost access, or instance misconfiguration — the default forced-Login V2 deployment is affected. Existing passwords, passkeys, or second factors do not prevent the reset, since completing it sets a new password and yields a live session. Deployments that do not run Login V2 (for example Login V1 only) are not affected, as Login V1 does not build these links from the public-host headers.
Scope note: This affects the Login V2 UI only. Exposure is instance-wide — any user on the instance can be targeted, across organizations, but not across instances. It is the login-side host-poisoning counterpart to GHSA-mwmh-7px9-4c23 , which addressed the same class of password-reset link poisoning through the Forwarded and X-Forwarded-Host headers; this issue reaches the same outcome through the public-host headers that Login V2 trusts, which were not covered by that fix.
Systems running one of the following versions are affected:
4.x: 4.0.0 through 4.19.4 (including RC versions)
The vulnerability has been addressed in the latest releases.
The patched version validates the public host taken from request headers against the instance’s trusted domains before it is used, and rejects a request whose public-host header is present but does not resolve to a trusted domain instead of silently falling back to an empty host. As a result, the host embedded in password-reset, invite, and email-verification links can no longer be set to an untrusted domain by a request header.
4.x : Upgrade to $\ge$ 4.19.5
The recommended solution is to upgrade to a patched version. There is no configuration that fully closes this gap on unpatched versions.
As a partial mitigation, a ZITADEL fronting proxy can be configured to strip or overwrite the x-zitadel-public-host header (along with x-zitadel-forward-host and x-forwarded-host ) on all inbound requests, so that the public host cannot be set by a client.
If you have any questions or this advisory, please email us at security@zitadel.com
Thanks to Lucas Dodgson from InfoGuard Labs ( @lucasdodgson ) for finding and reporting the vulnerability.
The full story
This article is one source in a clustered incident — the cluster page carries the summary, timeline and every other outlet covering it.
