Intervention

Intervention is a deliberate action that changes the course of an autonomous process. It can pause, redirect, reject, roll back, contain, or escalate execution.

Intervention is not a failure of autonomy. It is part of how autonomy remains governable.

Intervention Types

Veto

A veto prevents an action from proceeding. It may be triggered by policy violation, missing authority, insufficient context, high risk, or conflict with system intent.

Hold

A hold pauses execution until more information, review, or authority is available.

Escalation

Escalation routes a decision or exception to a person, team, policy owner, or higher-level control plane.

Human Review

Human review is useful when judgement, accountability, ambiguity, legal exposure, or organisational context cannot be delegated safely to automation.

Automated Policy Check

An automated policy check evaluates rules, thresholds, permissions, or risk conditions before allowing execution to continue.

Rollback Trigger

A rollback trigger reverses or compensates for a change when outcomes move outside acceptable bounds.

Containment

Containment limits the blast radius of a process by restricting tools, scope, budget, state access, or downstream actions.

Designing For Intervention

Intervention works best when it is designed before it is needed. A system should know what can be paused, who can approve continuation, what evidence is required, and how the intervention is recorded.

An autonomous system with no practical intervention path is not fully governable.