GPU VulnDB

Database/Kernel, userspace & hypervisor

OpenSSH ssh client: control characters in untrusted usernames can execute commands via ProxyCommand

CVSS 3.6CVE-2025-61984Kernel, userspace & hypervisorcurated

Impact

The ssh client before 10.1 accepts control characters in usernames that come from the command line or from %-sequence expansion in a config file, and those characters can survive into a ProxyCommand, turning a hostile username into shell command execution on the machine running ssh. On a GPU fleet this matters wherever ssh invocations are built from data an operator does not fully control: job scripts that template a user into a jump-host ProxyCommand, Slurm prologs, orchestration glue that shells out to ssh with an attacker-influenced target. Execution lands in the context of whatever account drives the automation, which on a management or login host is often broadly privileged across the cluster. Fleets that use plain literal usernames in ssh_config and never pass usernames in from outside are not exposed.

Who can reach it

Local, and only through a path that feeds an attacker-chosen username into ssh on the command line or into a %-expanded config value, with a ProxyCommand in effect. No authentication to the remote host is needed - the command runs before the connection is established. A config file that hardcodes a complete literal username is not an exposed source.

What to do

Upgrade OpenSSH to 10.1p1 or a distribution backport. The client is not a long-running daemon, so this is a package update with no node drain or reboot needed; already-running sessions are unaffected and new invocations pick up the fix. Until patched, stop interpolating untrusted usernames into ssh command lines and ProxyCommand templates. Siemens has published a product advisory for affected SIMATIC devices; follow it separately for embedded equipment.

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.