TechPastWeek logo
Security8 min read

SSHole OpenSSH Flaw Exploited Within Hours as CISA Sets 12 November Deadline for US Agencies

Attackers began exploiting an OpenSSH login bypass on musl-based Linux hours after the 5 November fix. CISA told US federal agencies to patch by 12 November.

The rear of a server rack in a data center, with network switches, patch panels and bundled Ethernet and power cabling.
On this page

Key takeaways

  • CVE-2026-123158, nicknamed SSHole, is an OpenSSH authentication bypass rated CVSS 8.1 and fixed in version 10.6p2 on 5 November 2026.
  • It affects musl-based Linux systems that do not check passwords through PAM, including Alpine's default openssh-server and many OpenWrt, Buildroot and Yocto builds.
  • Previdian observed the first attempt less than nine hours after the CVE was published and confirmed root access on an Alpine honeypot; GreyNoise recorded successful exploitation on 7 November at 00:46 UTC.
  • CISA added the vulnerability to its Known Exploited Vulnerabilities catalog on 9 November and set a 12 November deadline for US federal civilian agencies.
  • Fixing it requires OpenSSH 10.6p2 and a restart of sshd, because the package install leaves the running daemon unchanged.

OpenSSH 10.6p2 shipped on 5 November 2026 with a fix for an authentication bypass that lets an attacker log in over SSH without knowing a password, on any account that has no password hash, root included. Less than nine hours after the CVE was published, the first attempt arrived at a honeypot. By 7 November, GreyNoise had recorded successful exploitation in the wild. On 9 November, CISA added CVE-2026-123158 to its Known Exploited Vulnerabilities catalog and gave US federal civilian agencies until 12 November to patch.

Related reading: AI Agents Found a Zero-Day in MONAI Label and Rooted a University Server · China-Linked Hackers Exploited a vCenter Zero-Day Weeks Before the Patch.

The flaw was found by researcher Özgün Kültekin of Wakefield and nicknamed "SSHole". It carries a CVSS score of 8.1 and is classified as CWE-394, an unexpected status code or return value. The nickname is crude. The mechanism is a return value that musl and sshd read differently.

A fix, then exploitation within hours

OpenSSH published 10.6p2 on 5 November. Alpine Linux, whose default server build is one of the exposed configurations, shipped patched packages for every branch from 3.16 to edge the same day and published an advisory that described the consequences in plain terms. CISA took four days to add the CVE to its catalog of flaws known to be exploited. The entry gives the vendor as OpenBSD and the product as OpenSSH, with a due date of 12 November.

For anyone running SSH on a musl-based system, the order of those dates matters more than the score. A proof-of-concept exploit was available on 5 November, BleepingComputer reported, and successful exploitation followed the next day.

The bug takes one connection, and the conditions it needs, an account with no password hash and a reachable SSH port, are common on the systems that use musl.

Why a 300-character password logs you in

OpenSSH keeps password hashes in /etc/shadow. An account with * in place of a hash has no usable password. The affected accounts are those configured that way: root, key-only users created without a password, the alpine user in Alpine's official tiny-cloud images, and users from a directory service reached through musl-nscd. In each case the intent is that no password can match.

When sshd is asked to authenticate against one of those accounts, it does not stop at the missing hash. It borrows the salt from another local account that does have a real hash, then hashes the supplied password anyway. OpenSSH has done this since 2016, when the behavior was added to close a username-enumeration timing leak tracked as CVE-2016-6210. The constant work keeps response times even, so an attacker cannot learn which usernames exist from how long a failed login takes.

musl changes the result. Since 2012, its crypt() has refused SHA-256 and SHA-512 password inputs longer than 256 bytes and returned * on failure. sshd compares that output with the * stored for the passwordless account. The strings match, and the login succeeds. A password between 257 and 1024 characters is enough.

Three conditions have to line up: password authentication must be enabled, which is the OpenSSH default; the target account must have * in /etc/shadow; and another local account must have a real password hash for sshd to borrow the salt from. The attacker does not need to know that other password. A machine with no password hashes at all would not supply the salt, which is one reason the flaw is not universal.

Alpine's advisory is direct about the result:

"An attacker who can reach your SSH port can log in without knowing any password. It takes one connection, works every time, and needs no retries." — Alpine Linux security advisory, 5 November 2026

Who is actually exposed

The vulnerable set is narrower than "OpenSSH". It needs musl, no PAM check for passwords, a passwordless target account and password authentication left on. Debian, Ubuntu and Fedora build against glibc and are not affected, Kültekin wrote. Neither are systems that check passwords through PAM, nor Dropbear, nor OpenBSD's own sshd.

Alpine is the obvious case. Its default openssh-server package is built with musl and without PAM for password checks. OpenWrt, Buildroot and Yocto images that use musl fall into the same group, as do many container images. Kültekin's write-up points readers at "a lot of the docker images you've pulled".

Alpine patched every supported branch on 5 November. SOCRadar listed the affected tiny-cloud images at versions 3.18.11 through 3.24.3. If you run Alpine on a VPS, in a container or as the base for appliance firmware, the version check is worth doing even when the box sits behind a firewall. The flaw needs one reachable SSH port, not a whole exposed network.

What the attackers did

Previdian, Ryan Dewhurst's security firm, runs honeypots. Its first observation came less than nine hours after the CVE was published, and it confirmed root access on a honeypot running real Alpine Linux and OpenSSH. The source address, 89.116.89[.]18, sent oversized passwords and then ran id, uname and hostname through non-interactive SSH sessions. In total, Previdian counted 20 attempts from 13 unique addresses in nine countries, first seen on 6 November and last on 8 November.

GreyNoise recorded its first successful exploitation on 7 November at 00:46 UTC, about 40 hours after the fix. It logged 16 successful exploitation chains from six source addresses tied to post-exploitation activity. The intruders checked system and container details, queried cloud metadata services, tried to read password hashes and root SSH keys, and tried to add their own keys to authorized_keys for persistence. One got root and installed "TAC Agent", a Python command-and-control implant. The implant could not reach its server at tac.vpnla[.]org and never received a task. Its comments, logs and error messages were in Chinese.

That last detail is a clue, not an attribution. GreyNoise says its honeypot data is not representative of broader activity, and it recorded 27.5 million scans of port 22 and 20.1 million attempted SSH sessions in the same period. Those scans are background noise, not SSHole attacks. The counts that matter are smaller: a few dozen attempts and a handful of confirmed chains, not a victim tally.

WatchTowr said on 6 November that it had seen the flaw exploited in attacks linked to botnets and cryptomining. SecurityWeek reported the same pattern and said other threat actors were likely to follow. Defused described "a small amount of actors" attempting exploitation. No named victim, no confirmed widespread campaign and no reported data theft have emerged from any of it.

The speed is explained by the exploit's simplicity, not by a sophisticated campaign. One connection with a long password is enough. A botnet can try that against every address it scans.

CISA's three-day deadline

CISA published catalog version 2026.11.09 at 19:54 UTC on 9 November. The SSHole entry carries a dateAdded of 9 November and a dueDate of 12 November, three days later. It lists ransomware use as unknown, requires forensic triage and tells agencies to apply mitigations per vendor instructions and BOD 26-04, then evaluate their internet exposure.

The deadline is not a private-sector mandate

The 12 November date binds US federal civilian agencies covered by BOD 26-04. It does not create an obligation for private companies, state governments or agencies outside the US executive branch. Those operators should still treat the flaw as urgent, because the exploit is public and already in use.

The listing is a public signal as much as a legal one. It tells defenders the flaw is being used. That is different from a warning that it might be. Help Net Security noted on 6 November that CISA had not yet added the CVE to the catalog. That changed three days later.

For federal agencies, three days is the whole window. The work is more than applying a package: agencies must confirm which of their internet-facing hosts run a musl build without PAM, restart the daemon and check logs for logins that should not exist.

What to do this week

Start with the version. OpenSSH 10.6p2 is the fixed release. Alpine users should have patched packages from 5 November for branches 3.16 through edge. Updating the package is half the job: installing it leaves the running sshd untouched. Restart the daemon, or reboot, then confirm the process is running the new binary.

Then remove the condition attackers need most. Setting PasswordAuthentication no closes the password path and is the right default for servers that authenticate with keys.

Hunt for evidence:

  • Search /var/log/messages for "Accepted password for" entries naming accounts that should have no password: alpine, root, key-only users.
  • Review use of doas, recently created users and changes to authorized_keys files.
  • Check whether the host is reachable on port 22 from the internet at all.

If a login cannot be ruled out, treat the machine as compromised and rotate the keys and passwords it could reach. Exposed credentials can stay useful to an intruder long after the breach that leaked them; for a recent example, see our report on the Chainalysis breach. The implant GreyNoise saw was a Python script that never phoned home, but the next one may work.

Be careful with proof-of-concept code from GitHub. Previdian warned that the exploits circulating there are unverified and may contain malware.

The list of unaffected systems is worth keeping in view. glibc distributions, PAM-based password checks, Dropbear and OpenBSD do not have this bug. Many of the people who need to act are running musl because it came with the base image.

A patched package on disk and an old sshd in memory is still a vulnerable server. Check the process.

Tags

opensshssholecisa-kevactive-exploitationalpine-linuxauthentication-bypass

Frequently asked questions

What is SSHole (CVE-2026-123158)?
SSHole is an authentication bypass in OpenSSH, tracked as CVE-2026-123158 and rated CVSS 8.1. On vulnerable musl-based systems, a password of 257 to 1024 characters can be accepted for an account that has no password hash, including root.
Is my Linux server affected by the OpenSSH flaw?
Only if it runs OpenSSH before 10.6p2 on musl without PAM password checks, has password authentication enabled, and has a target account with an asterisk in place of a password hash. Debian, Ubuntu and Fedora use glibc and are not affected, and neither are systems that check passwords through PAM.
What is the CISA deadline for CVE-2026-123158?
CISA added the flaw to its Known Exploited Vulnerabilities catalog on 9 November 2026 and set a due date of 12 November 2026. The deadline binds US federal civilian agencies under BOD 26-04, not private companies.
How can I tell whether the flaw was exploited on my system?
Search /var/log/messages for "Accepted password for" entries naming accounts that should have no password, such as alpine, root or key-only users. Also review doas use, recently created users and authorized_keys changes. If a login cannot be ruled out, treat the machine as compromised.
Share

Found this useful? Share it with your team.

TPW Editorial

TPW Editorial

Editorial Team

TechPastWeek's editorial team reports and edits every story we publish. We research each piece against primary and credible secondary sources, and a human editor reviews and signs off before it goes live.

Keep reading