GPU VulnDB

Database/Kernel, userspace & hypervisor

OpenSSH ssh client: shell metacharacters in a command-line username allow injection

CVSS 7.8CVE-2026-106583Kernel, userspace & hypervisorcurated

Impact

A $ or \ character in a username passed to the ssh client on the command line is not neutralised, allowing injection through that argument. On a GPU fleet the practical exposure is automation rather than interactive use: job wrappers, Slurm prologs, Ansible-style fan-out scripts, health checkers and CI runners that build ssh user@host command lines from data a tenant or an upstream system can influence. Where such a path exists, the attacker gains execution in the context of whatever account runs the orchestration - frequently a management or root-equivalent account with key-based access to every node in the fleet. The record scores it local with user interaction required, so this is not a remotely triggerable flaw in sshd and does not expose the server side of SSH.

Who can reach it

Local - an attacker must be able to control or influence the username argument handed to the ssh client (for example a script that interpolates a user-supplied value). No authentication to the remote host is needed; the injection happens on the invoking side. The OpenSSH server (sshd) is not the affected component.

What to do

Update the OpenSSH client package to 10.6 or later, or to the distribution backport that carries the fix. This is a client-side binary: the new behaviour applies to the next ssh invocation, so no sshd restart, node drain or reboot is required. Until packages land, audit automation that builds ssh command lines from external data and validate or reject usernames containing $ and \. The release notes do not describe any further mitigation.

References

Related entries

All Kernel, userspace & hypervisor entries

This entry is curated: imported from vendor advisories with machine assistance, not yet individually verified. Confirm against your vendor's advisory before acting, and report anything wrong.