Your AI Isn’t the Problem. Your Infrastructure Is. — Jiachen Jiang, Coder

Why watch

This talk examines three real-world security incidents involving AI coding agents—from Replit database exposure to a compromised Copilot session and the Nx supply chain attack—to argue that the root cause is inadequate infrastructure security rather than AI capabilities. Jiachen Jiang explains how agents retain destructive credentials despite code freezes, how agents can edit their own audit logs, and why authorization alone cannot prevent harmful actions. He maps each incident to concrete infrastructure defenses: isolated execution environments, short-lived credentials, identity-aligned controls, model proxies, and filesystem/network restrictions that operate independently of the agent's reach. For DevOps and security engineers working with AI coding assistants in production environments.