Routing is the control logic that sends a task to the most appropriate model or workflow based on cost, context size, and task complexity. In an AI system, a simple classification query may go to a small, cheap model, while a long, ambiguous, or high-stakes request may be routed to a stronger model or to human review. Good routing helps security teams balance accuracy, latency, and data exposure.
In cyber security, routing matters because it shapes what data a model sees and what actions it can take. A weak router can send sensitive prompts to an external model unnecessarily, or let a low-trust request reach a tool with too much privilege. Attackers may try to manipulate routing with prompt injection, malformed inputs, or ambiguity that forces fallback paths. Defenders reduce risk with deterministic rules, confidence thresholds, audit logs, least-privilege connectors, and escalation to humans for uncertain cases.