Run the v3.1 base equations over the vector both publishers agree on and it comes to 9.1 [1][2]. The two metrics carrying that number are PR:N and UI:N: the attacker needs no account, and the victim does nothing [1]. That squares with the defect as described, since the reset-credentials flow accepts new credentials without the email verification link being followed, which Red Hat attributes to improper state validation inside the flow and rates Critical [3][4]. The GitHub issue tracking it is titled, plainly, unauthenticated account takeover via reset-credentials flow bypass [15].
Worth noting what else travelled with it. The 26.7.2 security list covers seven issues and eight CVEs [6][7][8], one of the others being CVE-2026-15571, a predictable account-linking hash that also yields account takeover through a malicious OIDC client [7]. Two separate takeover paths closed in one release, in the component that all the applications behind it delegate login to [9].
The awkward part sits after the upgrade. In a Keycloak community forum thread, an operator found that offline refresh tokens kept issuing access tokens after the user's password was reset, and a respondent confirmed that an attacker refreshing before the legitimate user could do so indefinitely [11]. The suggested route was a custom EventListenerProvider on the UPDATE_PASSWORD event; the poster reported that removeUserSessions handled sessions but that revoking offline tokens from inside a listener still did not work for them [12]. So a forced reset, the obvious clean-up after a bypass like this, is not by itself proof that the account is back.