[ctx]ctxseccontext security: attacks on the harness and context layer of AI agents, plus exploited-now news, hunting and playbooks.

← Tips & Guides · markdown

tip based on the Tailworm report's interim mitigations (Oct 25); vendor fix pending

Tip: until Tailscale ships a fix, treat every Tailscale SSH rule as root access and narrow it

The Tailworm findings only work through SSH rules you already granted, so the policy file is where to act.

  • Find which nodes run Tailscale SSH (tailscale debug prefs shows RunSSH) and switch it off with tailscale set --ssh=false where you don't use it.
  • Replace rules like "group:eng may SSH to tag:server" with specific source-to-destination pairs.
  • Don't allow root or passwordless-sudo accounts as SSH targets, and disable forwarding on each rule unless someone needs it. On headscale, forwarding is always on, so treat any SSH grant as root-equivalent.
  • Remove tailscale serve --tcp forwards that point at unix sockets, especially the agent's own socket.
  • If a node may have been compromised, delete it in the admin console. Reimaging alone doesn't revoke a copied node key.

Sources: Cyberdione Labs (Erica Windisch): Tailworm security assessment of Tailscale v1.104.2 and headscale v0.29.5 (Oct 24-25) · Tailscale security bulletins (latest TS-2026-013, Oct 21; nothing for Tailworm yet)