
CVE-2026-123158: OpenSSH's musl Password Bypass Is in CISA's KEV — and the Patch Deadline Already Passed
Any password between 257 and 1024 bytes could log into an account that had none, because of how musl's crypt() and OpenSSH's salt-borrowing logic interact. GreyNoise confirmed root compromises within about 40 hours of disclosure.
Overview
CVE-2026-123158 is an authentication bypass in sshd, the server component of OpenSSH, on Linux systems built against musl libc — Alpine Linux's default C library — without PAM. The CVE record describes it as a bug "related to misinterpretation of whether a string is intended to be a password hash": for some account configurations, any password between 256 and 1024 characters succeeds, regardless of the account's real password. OpenBSD shipped the fix, OpenSSH 10.6p2, on November 5, 2026; Alpine shipped a patched package the same day.
- Advisory date: November 5, 2026 (OpenSSH 10.6p2 release and Alpine Linux advisory)
- CVSS: 8.1 HIGH — AV:N/AC:H/PR:N/UI:N/S:U/C:H/I:H/A:H
- CWE-394: Unexpected Status Code or Return Value
- KEV status: added November 9, 2026; federal remediation deadline November 12, 2026 — already passed
- Affected: OpenSSH before 10.6p2, on Linux, built against musl libc, with PAM disabled (UsePAM no, or a non-PAM build) and PasswordAuthentication yes
Scores as of 2026-11-13live record →
| Alpine branch | Fixed openssh package | Fixed release bundle |
|---|---|---|
| edge | 10.6_p2-r0 | — |
| 3.24 | 10.3_p1-r2 | 3.24.4 |
| 3.23 | 10.2_p1-r1 | 3.23.8 |
| 3.22 | 10.0_p1-r11 | 3.22.8 |
| 3.21 | 9.9_p2-r1 | 3.21.10 |
| 3.16 – 3.20 (unsupported) | various backports, 9.0–9.7 | not bundled — move to a supported branch |
How a C library error code became a password
The chain starts in musl, not in OpenSSH. musl's crypt() function returns the literal string "" as an error indicator whenever the key it's given is longer than 256 bytes. Separately, every Linux account with no password set also stores "" in its /etc/shadow password field — the standard convention for "no password, can't log in this way." sshd's non-PAM authentication path needs a real salt to run crypt() against in the first place, so for an account whose own field is too short to contain one, it borrows the salt from any other account on the system that does have a real $5$/$6$ hash. Send a password between 257 and 1024 bytes to the passwordless account, and crypt() returns "" — which matches the "" already stored there. Alpine's own summary of the bug: "One user with a password is enough: it makes every account without one reachable."
Why an oversized password passes sshd's check
- Attacker sends a 257–1024 byte password — Target: an account with no password set — shadow field "*" — e.g. a key-only user
- sshd borrows a salt — Non-PAM sshd needs a real $5$/$6$ salt to run crypt() — it borrows one from any other account on the system that has a real password hash
- musl crypt() rejects the oversized key — musl's crypt() returns the literal string "*" as its own error indicator for any key over 256 bytes
- "*" equals "*" — The borrowed-salt crypt() output matches the stored "*" password field exactly
- Login succeeds — sshd grants access to the passwordless account — no real password was ever set or needed
A fix with no embargo window
Most CVEs get weeks of private coordination before anyone outside the vendor and reporter knows a bug exists. This one barely got a day.
CVE-2026-123158's first week
- Fix committed to openssh-portablePatch commit 4353fdc, per GreyNoise's published timeline of the disclosure.
- Public disclosureOpenSSH 10.6p2 released within a minute of the fix going public, per GreyNoise. Alpine Linux published its own advisory and patched packages the same day.
- CVE-2026-123158 publishedNVD/CVE.org record published at 15:46:26 UTC, CVSS 8.1. A public exploit reportedly appeared roughly 2.5 hours before the CVE ID itself was reserved, per GreyNoise.
- First confirmed root compromiseGreyNoise recorded successful exploitation on a honeypot running real Alpine Linux and OpenSSH — about 40 hours after public disclosure.
- Added to CISA KEVCatalogued as an authentication-bypass vulnerability; federal civilian agencies given until 2026-11-12 to remediate under BOD 26-04.
- KEV deadline passedThe three-day federal remediation window closed. We found no amendment to the KEV entry extending it.
We consider the root cause of this bug to be outside sshd, but it is exploitable via sshd and can be avoided there.
OpenSSH maintainer Damien Miller explained the unusually fast, nearly embargo-free turnaround in a mailing-list post the same day: the project now treats long private-disclosure windows as a risk in their own right, citing a trend of "rapid bug rediscovery by AI agents, sometimes within hours" as a reason delay increasingly helps attackers more than defenders. The report was credited to a researcher at Wakefield (wakefield.co), with the mailing list noting the same vulnerability "does not exist on non-Linux systems, Linux/glibc systems, and Linux systems that use PAM for password authentication."
What actually happened on the internet
GreyNoise has tracked scanning and exploitation traffic for this CVE since public disclosure. The overwhelming majority is noise — mass TCP/22 scanning that never gets near the actual bug — but a small number of connections turned into confirmed, logged intrusions. Weighed against the broader internet exposure picture we track, this is a case where raw scan volume was a terrible predictor of real compromise.
| Stage | Count | Share of prior stage |
|---|---|---|
| TCP/22 SYN scans | 66,000,000 | — |
| Full TCP connections | 48,400,000 | 73% |
| Attempted SSH logins | 47,500,000 | 98% |
| Possible exploitation attempts | 5 | <0.01% |
| Confirmed post-exploitation activity | 42 events / 10 attacker IPs | — |
GreyNoise groups part of that activity under an internal cluster label, "GreyNoise Threat 10" (GT10) — seven of the ten attacker IPs reusing infrastructure across intrusions. Victim systems GreyNoise observed span New Zealand, Laos, Australia, India, Japan, Hong Kong, South Korea, Taiwan and France. Attacker-side infrastructure includes addresses registered to Vodafone in Egypt, DIGI Spain Telecom, two separate Indonesian networks, Oracle Corporation in Singapore, and China Mobile. GreyNoise's tracker lists 520 indicators tied to confirmed exploitation and 123 files created or modified on victim systems; among its named examples, one Vodafone-Egypt address alone ran five separate intrusions totaling 132 shell sessions and 235 commands.
What the score says vs. what happened
- Attack Complexity: High (AC:H)
- Requires musl libc + PAM disabled + a second account with a real password hash on the box
- Privileges required: None · User interaction: None
- Base score 8.1 — High, one notch below Critical
- Alpine's own advisory: "one connection, works every time, and needs no retries"
- Two public PoC repos on GitHub, tagged Exploit by NVD's own reference list
- First confirmed root compromise ~40 hours after disclosure
- 42 confirmed intrusions across 10 IPs inside the first week
Mitigation guidance
- Upgrade OpenSSH to 10.6p2 or later, or your distro's backported equivalent — see the Alpine table above for exact package versions.
- Restart sshd after upgrading. On Alpine, "apk upgrade -U -a" alone does not replace the running daemon; the old, vulnerable process keeps serving connections until it's restarted.
- If you can't patch immediately, set "PasswordAuthentication no" and restart sshd. Systems that already disable password auth, or that run PAM with "UsePAM yes", are not affected by this specific bug.
- Audit for prior compromise: check /var/log/messages (or your syslog equivalent) for "Accepted password for" entries against accounts that were never supposed to have one.
- Cloud-init-provisioned images that already set PasswordAuthentication no by default are not affected, per Alpine's advisory — verify that setting rather than assuming it.
Updates
2026-11-13 — The CISA KEV deadline of November 12 has passed. We found no amendment to the KEV entry extending it, and no indication CISA has removed or downgraded it. If that changes, we'll update this section rather than publish a new article.
Does this affect regular Ubuntu, Debian, or RHEL servers?
Is Alpine's default cloud image vulnerable out of the box?
Do I need PAM enabled to be safe?
Has this been confirmed exploited, or just scanned for?
Data as of 2026-11-13live record →
Sources: OpenSSH 10.6p2 release notes, Alpine Linux security advisory, openssh-unix-dev mailing list, Damien Miller, Nov 5 2026, CVE.org record, NVD entry, CISA Known Exploited Vulnerabilities catalog, GreyNoise CVE-2026-123158 tracker.