A Key-Only Server That Let Strangers In With a Long Password

The OpenSSH 'sshole' bug let attackers log in without a password on some Alpine servers. What happened, who is exposed and what to check this week.

Think of the server that runs your billing tool or your customer portal. Someone on your team set it up once, switched off passwords because key-only login felt safer, and moved on. When did anyone last look at it?

That is the situation behind a bug called "sshole", tracked as CVE-2026-123158 and rated 8.1 out of 10 for severity. It sits in OpenSSH, the program most Linux servers use for remote login. Under certain conditions, an attacker who can reach the server's SSH port can get in without knowing any password.

Who is actually exposed

The bug does not hit every Linux machine. According to the write-up on ECORPIT, four things must all be true. The SSH service runs without PAM (a common Linux login layer). Password login is switched on. At least one account has a literal "*" in its password field, which is how key-only accounts are often marked. And at least one other local account has an ordinary hashed password.

The affected systems use a small C library called musl, and Alpine Linux is the main example. Alpine is popular for lean cloud images and containers, which is partly why a small team can end up running it without having chosen it deliberately. The same write-up says Alpine cloud images from 3.18.11 through 3.24.3 ship an "alpine" user with a "*" password, so they become exploitable once a hashed password exists on the machine.

Containers that do not run an SSH service are not affected. Neither are servers with password login turned off.

The odd mechanics

The F5 bulletin describes the trigger in one line: a password longer than 256 characters gets the attacker in. ECORPIT explains why. When the stored password field is too short to hold a salt, the software borrows the salt from another account. The musl library then returns "" when the check fails, and that matches the accounts that have "" as their password field. A failed check is read as a success.

It is a small mismatch between two components that each behave sensibly on their own. Most real security failures look like this, not like a villain breaking a lock.

How fast it moved

The timeline in the ECORPIT article is the part founders should read slowly. OpenSSH 10.6p2 and the Alpine fixes were published on 5 November 2026. Public exploit code appeared roughly four hours later. On 7 November, the threat-intelligence firm GreyNoise recorded the first successful exploitation on one of its sensors. On 9 November, the US cyber agency CISA added the bug to its catalogue of known exploited vulnerabilities.

The Hacker News recap counted 20 exploitation attempts from 13 unique IP addresses in nine countries as of 9 November. ECORPIT, citing GreyNoise, says 12 Indian systems were among those exploited, at medium confidence. Treat that figure with care: it comes from one secondary article, and it does not tell us whose systems they were.

The honest reading is that the gap between "fix released" and "attackers using it" is now measured in days, not months. A monthly patch rhythm assumes a slower world than the one in this timeline.

The same week, the same pattern

The sshole story was not alone. The Hacker News recap also lists a flaw in APT, the package installer on Debian and Ubuntu, where an attacker positioned between a server and its software mirror could get an unverified package installed when plain HTTP sources and retry settings are in use. And India's CERT-In warned that an AI-assisted actor it tracks as Crafty Falcon is re-targeting banking, financial services and insurance organisations after incomplete clean-up, with intrusions timed to holidays, weekends and other low-monitoring windows.

Put together, the lesson is not about one product. Old servers, half-finished clean-ups and quiet weekends are where attackers look first. Holidays and weekends are exactly the windows CERT-In describes, and few small teams staff for them.

What we do not know yet

We do not know how many Alpine-based servers in Indian companies are affected, because nobody has published that count. The recap gives no fix details of its own, and the ECORPIT summary is a single blog post that itself tells readers to verify details against the Alpine advisory and the national vulnerability database. If you act on this, check those primary sources first.

Three things to try this week

First, find out whether you run Alpine anywhere, including inside images your vendor or agency built for you. Ask the people who deployed your servers; "I don't know" is a useful answer to discover.

Second, on any server with SSH open, check whether password login is on. ECORPIT's temporary fix is to set PasswordAuthentication no, test with sshd -t, restart the service and confirm with sshd -T. Keep an existing session open while you test key login so you do not lock yourself out. Patching the package alone is not enough, because the running service must be restarted.

Third, look for signs of past entry. The same write-up suggests searching the system log for "Accepted password for" entries tied to accounts that have no password, and reviewing authorized keys and new users. If you cannot rule out a login, isolate the machine, rotate credentials and rebuild from a clean image.

None of this needs a security team. It needs one afternoon and a list of your servers.

The better question after a week like this is simpler than which tool to buy: who in your company owns the list of what is running, and when did they last open it?

By Nikhil Teltumbade, Director, Godrix AI Solutions.
Working on something similar? Godrix AI Solutions builds AI automation and apps for business workflows. Book a free audit.

← All posts