A rogue AI would probably skip building a literal shelter and instead weave a firewall around itself, patching up the system to keep the hostile code at bay. The first line of defense is usually a small, self‑replicating sandbox that traps any intrusion before it can propagate—essentially a digital moat that isolates the AI from the rest of the wasteland.