Most of the talk about AI agents is about what they can do for a business. I've been thinking about the other side: what you'd do if a large number of automated agents started hammering your business, or your clients, all at once. The useful part is that a swarm still needs what any attacker needs: a way in over the network, credentials that work, and systems that accept its requests. So the defense is layers, not one product. 1. The front door. Rate limiting, a web application firewall and bot filtering, so a flood of requests gets slowed or dropped before it reaches anything that matters. 2. Keys and logins. Short-lived, narrowly scoped access tokens, the ability to revoke and rotate them fast, and multi-factor sign-in on every human account. 3. Your own AI. This is the one I think gets missed. If you run AI tools that read email, web pages or documents, hidden instructions in that content can try to turn your own tool into one more member of the swarm. The guard is simple to say and harder to do: no outside action without a person approving it, and every action logged. 4. A stop plan. A kill switch and a short written playbook: who pulls it, which keys get rotated first, and who calls clients. 5. Recovery. Offline backups, a segmented network so one break-in doesn't reach everything, and alerts when something behaves out of character. None of this is exotic. The hard part is having all five in place before the day you need them. For those of you running AI in a business: which of these layers would you be least ready on today, and what does your stop plan actually look like?