On November 5, OpenSSH shipped 10.6p2 to fix a bug that lets someone log in to certain accounts with a long, made-up password. On November 9, CISA put it in its Known Exploited Vulnerabilities catalog. It has a CVE number (CVE-2026-123158) and a nickname (*sshole), and attackers have been trying it on real servers since at least November 6.
Most small companies won't be exposed. The bug needs a specific setup. But that setup is normal on Alpine Linux, and Alpine is what a lot of container images are built on. So it's worth ten minutes to check.
What the bug is
The flaw is in how OpenSSH's server, sshd, checks a password when it is built without PAM and runs on musl libc, the C library Alpine uses instead of glibc. The OpenSSH announcement on oss-security says a password longer than 257 characters makes musl's crypt() return the string *. Accounts with a locked password have * as their stored hash, so sshd sees a match and lets the login through. The reporter was Özgün Kültekin of Wakefield.
A follow-up on the musl mailing list narrows it down. All of these have to be true for an account to be at risk:
- sshd runs with
UsePAM noandPasswordAuthentication yes. - The target account has
*as its hash, meaning password login is supposed to be off for it. - At least one other account has a real password set.
- The first account with a password, in /etc/passwd order, uses a SHA-256 or SHA-512 hash.
That's narrow, but a small box with a locked service account and one user with a password can match it. Dropbear and other non-OpenSSH servers aren't affected, per the same post. The OpenSSH notes put the root cause in crypt(), not sshd, and musl hasn't merged its own fix yet. Its maintainer says he plans to change the failure value, which distributions could backport. One detail is disputed across write-ups: whether the cutoff is 256 or 257 bytes. It doesn't change what you do.
It is being exploited
CISA added it on November 9 as “OpenBSD OpenSSH Unexpected Status Code or Return Value Vulnerability.” The Hacker News reports that attempts began around November 6 and 7. GreyNoise saw post-exploitation from six source IPs across 18 attack chains: looking around, querying cloud metadata, trying to read password hashes and root SSH keys, and editing authorized_keys so the access survives a patch. watchTowr reported botnet and cryptocurrency-miner activity too. Those figures come from the press write-up, not from the vendors' own pages, so treat them as reported.
The authorized_keys part is the one to remember. If someone got in before you patched, upgrading doesn't kick them out.
Do you run this?
Alpine's OpenSSH package is built without PAM, which is why Alpine is the headline case (that part comes from vendor and press write-ups; I didn't find it stated in the OpenSSH announcement). Other musl-based systems can be affected too. What matters is whether sshd is running at all. Most application containers don't run sshd, and a container without it isn't exposed through this bug. The risky ones are small VMs, bastion hosts, appliances, and any container where someone added openssh-server years ago.
- Find sshd. On each Alpine host or image, check whether openssh-server is installed and running (
apk info openssh-server,rc-service sshd status). Check from the outside too: is port 22 open to the internet on anything you own? - Turn off password logins. Set
PasswordAuthentication noin/etc/ssh/sshd_configand use keys. That's the workaround the OpenSSH announcement gives, and it's the right setting with or without this bug. - Patch and restart. Upgrade to a build with the 10.6p2 fix (on Alpine,
apk upgrade -U -a), then restart sshd. A running daemon keeps the old code until it restarts. Distributions backport fixes, so check the package changelog rather than trusting thessh -Vnumber. - Rebuild images. If an image bakes in openssh-server, rebuild it from a patched base and redeploy. Patching a running container doesn't fix the image you start next week.
- Look for signs of entry. Check
authorized_keysfor every account, including ones that shouldn't have any. Review auth logs for successful logins from addresses you don't know, especially to locked accounts such as root or service users. If anything looks wrong, rebuild the host and rotate the keys and credentials it held.
Where an outside test helps
An external test can tell you whether SSH is reachable from the internet. That's the first question here. It can't see which C library the server uses or how sshd is configured inside, so the checks above are on you. An open port 22 you didn't know about is worth finding either way. For what that kind of test covers, see what a $495 pen test covers. For how fast attackers move after a disclosure, see four days from disclosure to exploit.