Most of the servers running OpenSSH right now are not affected by the bug that got the catchy name this month. That is worth saying at the top, because the headlines around CVE-2026-123158 make it sound like the end of secure shell as we know it. It isn't. The flaw, nicknamed SSHhole (and spelled less politely as *sshole by the people who found it), only bites a specific build of OpenSSH in a specific configuration. The problem is that the specific configuration is extremely common in the places that tend to be forgotten: container images, embedded devices, and the minimalist Linux distributions that power a lot of modern infrastructure.
On November 9, 2026, the US Cybersecurity and Infrastructure Security Agency added CVE-2026-123158 to its Known Exploited Vulnerabilities catalog and gave federal civilian agencies until November 12 to patch under Binding Operational Directive 26-04. CISA only adds a flaw to that list when it has evidence of real attacks, not theoretical ones. MITRE scores the bug 8.1 out of 10 on the CVSS 3.1 scale, which lands it firmly in the High range. OpenSSH shipped the fix, version 10.6p2, on November 5.
What the bug actually is
Start with two pieces of plumbing. OpenSSH is the software almost every Linux and Unix machine uses to accept remote logins over SSH. musl libc is a small, efficient implementation of the standard C library, the collection of low-level functions that programs lean on for things like string handling and password hashing. Alpine Linux, which is the base layer for a huge share of Docker containers, ships with musl instead of the more common glibc. PAM, short for Pluggable Authentication Modules, is a separate framework that many Linux systems use to make authentication decisions. SSHhole only appears when OpenSSH is built against musl and PAM is not in the picture.
Here is the mechanism, which is almost elegant in how small it is. When an account has no usable password set, the system records that fact by putting a single asterisk, the * character, in the password field. Key-only accounts do this all the time. You log in with an SSH key, never a password, so the password slot holds * to mean "nothing valid here." The intent is that no typed password can ever match, because no password hashes to a bare asterisk.
On a musl build without PAM, that assumption breaks. Researchers at Horizon3 describe it plainly: an attacker who knows the name of such an account sends an absurdly long password, and musl's handling turns that input into the same lone *, so the comparison succeeds. Analysts at Ox Security reported that for some account setups, any password between roughly 256 and 1,024 characters does the trick. The attacker never knew a secret. They just knew that the account existed and that the server would accept a long enough string of garbage as if it were the right answer.
No brute force. No stolen key. Just a long string and a library that blinked.
Two conditions have to hold. Password authentication must be enabled in the SSH daemon, and the target account has to be one with no password set. That second condition sounds like it would protect most systems, until you remember how many service accounts, deploy accounts, and automation identities are created key-only on purpose. Those are exactly the accounts a defender assumes can never be reached by a password guess.
Who is exposed and who can relax
The clean dividing line is the C library and PAM. If your OpenSSH runs on glibc, you are not affected. If PAM is doing the authentication, you are not affected. Dropbear, the other popular small SSH server, is not affected, and neither are OpenBSD systems. The exposure is concentrated in musl-without-PAM builds: Alpine Linux versions 3.16 through 3.24 and the edge branch in their default state, OpenWrt, and the long tail of embedded gear that quietly ships OpenSSH on musl.
That container angle is the part worth sitting with. A single vulnerable Alpine base image can be baked into thousands of running containers across an organization, and most teams have no quick inventory of which images were built with password authentication left on. The flaw is not interesting on a locked-down host that only accepts keys through PAM. It is interesting on the forgotten container with a key-only service account, SSH exposed inside the cluster network, and password auth never explicitly turned off because nobody thought it mattered.
Exploitation is not hypothetical. By November 9, The Hacker News reported roughly 20 exploitation attempts from 13 distinct IP addresses, with source traffic spread across Denmark, Iran, Japan, the Netherlands, Norway, Singapore, Sweden, Switzerland, and the United States. Those are early numbers from honeypot-style telemetry, not a measure of successful break-ins, but they tell you scanners are already looking.
It is also exactly the kind of flaw a penetration tester will reach for on the next engagement, because it needs so little. There is no payload to smuggle past a defense and no credential to crack. You enumerate account names, find one on a musl host with password authentication left on, and send a long string. For anyone running an internal assessment, the value is less in the login itself and more in what it reveals about assumptions: that the organization treated key-only accounts as unreachable and never audited which of its container images still accept passwords. That gap is the finding worth writing up, whether or not the single account you land on leads anywhere interesting.
What to do this week
The fix is OpenSSH 10.6p2. Alpine has backported it across its supported branches, with specific package versions for edge and for 3.21 through 3.24, and advisory notes covering the older 3.16 through 3.20 line. Patching the host is the real answer, but patching containers means rebuilding and redeploying every image that inherited a vulnerable Alpine layer, which is slower and easier to half-finish. Plan for the rebuild, not just the apk upgrade on one machine.
If you cannot patch everything immediately, the interim move is to set PasswordAuthentication no in the SSH daemon configuration and restart the daemon. That closes the door completely, because the attack requires password authentication to be enabled. On a musl system you can alternatively bring PAM into the authentication path, which also sidesteps the flaw. Both are stopgaps. The patch is the fix.
What I would check first is not the version string but the behavior. Pull your authentication logs and look for successful password logins on accounts that are supposed to be key-only. A service account that has never once logged in with a password, suddenly accepting one, is the signal that matters here, and it is more reliable than trying to enumerate every image build flag from memory. If you run internet-facing SSH on anything that might be musl-based, treat that log review as the priority and the version audit as the follow-up.
SSHhole is a good reminder that "no password set" is a configuration, not a wall. The asterisk in a password field is a convention that depends on the library underneath honoring it. When the library quietly disagrees, a design that looked airtight for years turns into a one-request login. The lesson for defenders is the boring, durable one: know which of your systems run musl, know where password authentication is still enabled, and do not assume a key-only account can never answer a password prompt.
References
Horizon3.ai. (2026, November). CVE-2026-123158: OpenSSH musl bypass (*sshole). Horizon3.ai. https://horizon3.ai/attack-research/vulnerabilities/cve-2026-123158/
Kumar, A. (2026, November 9). Weekly recap: SSHole exploited, APT root, and more. The Hacker News. https://thehackernews.com/2026/11/weekly-recap-sshole-exploited-apt-root.html
eSecurity Planet. (2026, November). CISA adds four exploited flaws to KEV, sets Nov. 12 patch deadline. eSecurity Planet. https://www.esecurityplanet.com/vulnerabilities/news-cisa-kev-gpustack-freeswitch-openssh-ibm-november-2026/