Why Stolen Passwords Still Need a Reliable Second Factor
Why Stolen Passwords Still Need a Reliable Second Factor The premise behind CVE-2026-107145 CVE-2026-107145 begins with a familiar event: an attacker knows a user's password. The affected twofactor_gauthenti
Why Stolen Passwords Still Need a Reliable Second Factor
The premise behind CVE-2026-107145
CVE-2026-107145 begins with a familiar event: an attacker knows a user's password. The affected twofactor_gauthenticator plugin for Roundcube checks a six-digit TOTP code after the password step, but vulnerable versions do not restrict repeated attempts well enough.
The CVE is not a claim that the plugin can be bypassed without credentials. It is a claim that the second factor becomes weak after the first factor is lost. That distinction is central to building a realistic response.
What MFA promises
Multi-factor authentication reduces dependence on one secret. A password can be reused, phished, or recovered from a breach. A TOTP code is short-lived and should not be guessable through unlimited requests. The second factor is useful only when the verifier enforces the assumptions around its use.
Rate limiting is one such assumption. A login service needs a durable record of failed verification attempts. That record must survive a new session, a logout, and requests sent through a different route. Otherwise the attacker can reset the attempt budget by changing connection state.
The upstream fix uses a persistent per-IP and per-username counter, incremental waits from one second up to ten minutes, and a block window. It also closes a self-test route that could reveal whether a code was correct. These controls support the original security promise of the second factor.
Related bypass paths
CERT Polska disclosed CVE-2026-107146 for a session that survived failed verification and CVE-2026-107147 for a whitelist bypass through spoofed client headers. The three issues show how one weak assumption can undermine a credential even when the user has followed the MFA process.
The session issue matters because a redirect is not revocation. The whitelist issue matters because an HTTP header is not proof of network location. The upstream commit addresses both with server-side session destruction and a trusted-proxy-aware client-IP resolver.
Impact to the inbox
Mail is a recovery hub. Access can expose password-reset messages, contracts, invoices, personal data, and business conversations. A mailbox compromise can also change forwarding rules, filters, and account settings, or reach connected systems that accept email as an identity signal.
The reviewed advisory does not establish an active exploitation campaign. The impact discussion is therefore a consequence analysis based on the documented authentication bypass, not a claim that a particular organization was attacked.
Interpreting the score
The CNA assigned CVSS 4.0 a 6.3 Medium score with network attack vector, high complexity, no privileges required, and low impacts. The password prerequisite is part of the practical risk model. The score does not mean every affected mailbox has the same business importance.
Prioritize accounts that can reset other credentials, approve payments, administer infrastructure, or receive sensitive personal data. Those accounts may need a more aggressive response than an isolated mailbox with limited access and strong external controls.
Practical response
Patch the plugin to commit 4640889d71a25391d8814de826f39416bf3db08d or later. Confirm that failed attempts are counted across logins and that the AJAX route honors the block. Verify server-side session destruction and proxy trust with tests that do not depend on the browser following a redirect.
Reset passwords for accounts suspected of compromise and revoke sessions in every system that shares the mailbox identity. Review mailbox rules, forwarding, connected applications, and recent security notifications. Preserve logs for the investigation.
Affected products
All plugin builds before commit 4640889 are listed as vulnerable. A modern Roundcube core does not imply a modern plugin. Inventory the plugin package and enabled configuration on each web node.
Exposure context
ZoomEye Search app="Roundcube" returned 61,971 product matches; the exact CVE Filter returned 0. The product count can guide webmail inventory, but it does not prove the presence of the plugin, the absence of the fix, or an active account takeover.
References
Originally published by Dev.to Security. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.