Tailscale SSH Vulnerabilities Allow Attackers to Gain Root Access to Connected Servers

Security researchers have revealed multiple high-severity vulnerabilities in Tailscale and Headscale that could let attackers gain root access to connected servers, bypass network security policies, steal credentials, and disrupt remote-access infrastructure.

Erica Windisch of Cyberdione Labs published these findings in a security research report titled Tailworm. The assessment examined Tailscale version 1.104.2 and headscale version 0.29.5 between October 22 and October 24, 2026.

Researchers demonstrated that an attacker controlling a single device could potentially compromise other servers accessible through Tailscale SSH without needing additional passwords, SSH keys, or administrator privileges.

These attacks rely on existing SSH access permissions and port-forwarding capabilities. Researchers reproduced the vulnerabilities in controlled laboratory environments rather than production networks.

Tailscale SSH Vulnerabilities

The main security issue lies in how Tailscale SSH manages Unix-domain socket forwarding through its privileged tailscaled process. Unlike traditional OpenSSH deployments that utilize privilege separation, Tailscale’s embedded SSH server performs certain forwarding operations with root privileges.

Researchers identified a vulnerability tracked internally as T3. This issue arises when local Unix socket forwarding with the command `ssh -L` allows an authenticated peer to access sockets that are normally restricted to root.

During testing, researchers exploited this behavior to communicate with a privileged Docker socket and execute commands with root permissions. They successfully accessed `/etc/shadow`, which contains sensitive Linux password information.

Another vulnerability, designated T2, involves remote socket forwarding via the `ssh -R` command. This vulnerability allows the privileged process to replace root-owned Unix sockets with attacker-controlled listeners, potentially enabling command execution, service impersonation, and interception of Docker registry credentials.

Additionally, a related time-of-check-to-time-of-use (TOCTOU) race condition lets attackers bypass file-path restrictions meant to protect sensitive system resources.

The report also illustrates how attackers could automate lateral movement across connected systems. By exploiting trusted device identities and permissive SSH access rules, attackers could move between accessible machines without requiring further user interaction.

The researchers demonstrated that propagation could occur in less than one second per hop under laboratory conditions.

Further findings reveal concerns regarding Tailscale’s experimental conn25 app connector, which could allow authorized devices to access unintended internal services, cloud metadata endpoints, and Kubernetes resources.

The assessment also uncovered memory-exhaustion vulnerabilities affecting both headscale and Tailscale components, which could disrupt device registration and network availability.

Administrators should immediately review Tailscale SSH permissions, restrict connections to specific source and destination devices, and turn off unnecessary port-forwarding capabilities. Organizations using Headscale version 0.29.5 face additional concerns, as this version does not support disabling SSH forwarding independently.

Researchers recommend limiting app connector permissions, enforcing destination allowlists, monitoring suspicious device enrollments, and protecting device credentials through hardware-backed security measures.

As of the report on October 25, no CVE identifiers or fixed versions have been announced for these vulnerabilities. Tailscale has acknowledged the research and is currently investigating the identified issues.

This discovery highlights how weaknesses in privileged SSH operations can turn a compromised device into a gateway for broader server compromises, even within networks designed around zero-trust security principles.

Stops Cyber threats before impact with 21 min faster MTTR. Integrate ANYRUN’s Sandbox in your SOC.