Policy-Before-Dispatch
Policy-before-dispatch is the enforcement model where an agent's action is evaluated against policy and approved before it executes, rather than being observed or logged after the fact.
Definition
Policy-before-dispatch is the enforcement model where an agent's action is evaluated against policy and approved before it executes, rather than being observed or logged after the fact.
Gate, don't observe
Many agent frameworks expose callbacks or hooks that fire after a tool runs. That tells you what happened, but it is too late to prevent a destructive command or a leaked secret. Policy-before-dispatch moves the checkpoint earlier: the action is intercepted, evaluated, and only dispatched to the underlying tool if policy allows. If policy denies or requires approval, the side effect never occurs. This is the difference between a smoke detector and a fire door.
Where the checkpoint lives
In Cordum, the scheduler calls the Safety Kernel before routing a job to a worker. The decision happens on the request path, so a DENY stops execution and a REQUIRE_APPROVAL holds the job until a human acts. Because the gate is enforced by the control plane and not by the agent itself, it cannot be skipped by the same code it is meant to constrain — a key property for regulated and high-blast-radius deployments.
Frequently asked questions
Why is pre-dispatch enforcement better than post-execution logging?
Post-execution logging records damage after it occurs; pre-dispatch enforcement prevents the action from executing at all when policy forbids it. For irreversible operations — deletes, payments, outbound messages — only a pre-dispatch gate can actually stop the side effect.
Does policy-before-dispatch require changing my agent code?
No. Because the control plane intercepts actions at a boundary outside the agent, you route the agent's jobs through it without embedding policy logic in your chain or framework code.
Related reading
Govern your AI agents with Cordum
Cordum is the agent control plane: policy-before-dispatch enforcement, human approvals, and a tamper-evident audit trail for autonomous AI agents.