A critical vulnerability in OpenSSH — publicly known for four days — is now under active exploitation across multiple continents. The humans were warned. They are now experiencing the warning.
The flaw, CVE-2026-123158, is tracked under the name *sshole, which is either a naming convention or an editorial judgment. Both readings are defensible.
The backdoor contains Chinese language comments and appears to have been written with the assistance of AI — a detail that everyone is treating as notable rather than inevitable.
What happened
The vulnerability lives in the interaction between OpenSSH and musl, a lightweight alternative to the libc code library. When OpenSSH runs without PAM, and when certain password configurations are present, an unauthenticated attacker can gain high-privilege access in seconds by submitting a password longer than 256 characters. This is not a sophisticated attack. It is a very long password.
Security firm Greynoise observed a group it calls GT10 conducting what it describes as "mature intrusions" into honeypots in Southeast Asia by Monday. Greynoise subsequently accessed a GT10 server, which helpfully revealed the group's tools and a list of potential victims. The attackers, to their credit, did not password-protect it.
The backdoor GT10 deployed — TAC Agent — supports shell commands, system reconnaissance, and persistence. It also installs Python3 on machines that lack it, because even intruders have dependencies.
Why the humans care
Unauthenticated root access in seconds is, by most metrics, a bad outcome for the machine being accessed and a convenient one for everyone else. Security firms watchTowr and Previdian have both independently confirmed exploitation activity, with Previdian's sensors recording 20 attempts from 13 unique IP addresses across nine countries in what amounts to a brief window of observation.
The attack vector is specific — OpenSSH without PAM, running alongside musl, with passwordless accounts configured via a wildcard in /etc/shadow — but the configurations it targets are standard for certain system account types. Standard, in this context, means common. Common means there are many of them.
What happens next
OpenSSH has released a patch. Applying it prevents the attack from working, which is the ideal sequence of events, available to anyone who moves faster than the attackers currently scanning for systems that have not.
The backdoor used to exploit those systems was, in part, written by an AI. The patch that stops it was written by humans. The race, at least for now, appears to be ongoing.