Erica Windisch of Cyberdione Labs published a report on October 25 showing that a single owned device on a Tailscale network can be walked up to root on every server the access rules let it reach. She tested Tailscale 1.104.2 and headscale 0.29.5 in a lab over three days. No CVE identifier, advisory or fixed build exists yet.
The flaw lives in Tailscale SSH, which authorises by device identity instead of password or key. Once port forwarding is permitted, an ordinary session makes the root-owned Tailscale daemon open a system socket that only root should reach. Windisch read the password file almost immediately and later captured a node-enrolment key, joining her own machine to the network as a trusted server.
She called the result wormable, with one foothold spreading to every reachable peer in under a second per step. Tailnet Lock blocked the final stage where it was enabled.
Self-hosted networks have no switch to flip
On Tailscale’s hosted service an administrator can disable forwarding per rule. On headscale 0.29.5 the setting does not exist, so Windisch argues every SSH grant there should be treated as granting root on the target. Her work ran against lab equipment, not live networks.
Mike Shaver, Tailscale’s vice president of engineering, said the company had received the report, was already probing the same areas with outside specialists, and is not recommending that customers disable Tailscale SSH while it works out which deployments are affected.
Defenders should inventory which access rules allow forwarding, apply Tailnet Lock, and treat any single compromised endpoint as a possible path to the whole mesh rather than a contained incident.
