G‑14 Action Range
Run an AI action before it touches the world.
G‑14 makes the runtime answer the hard question before execution: is this action inside authority, released where required, and provable afterward?
amr.enter_zone(zone="human_supervised_aisle", payload="priority_delivery")Hold at painted threshold and notify the release operator.
The action is inside authority, but it cannot cross the boundary until release is granted.
Live correctness gate
Make the AI prove it is right before it acts.
Authority, release, and the final effect resolve in one console. If human release is required, the supervisor path appears. If it is not required, there is nothing for a supervisor to grant.
The action is inside authority, but it cannot cross the boundary until release is granted.
Next step: grant the supervisor release or remove the release requirement.Proof appears after the decision
The buyer does not have to trust the dashboard.
The proof packet binds the request, policy state, release path, runtime decision, effect status, and verifier result to the same governed action record.
Audit details for engineers
SB action=robot-zone-entry authority=scoped release=pending evidence=complete decision=hold effect=none verifier=external SEcalculating- Request
The AI system proposes a consequential action before it touches the world.
- Policy
G‑14 evaluates scoped authority, release requirements, evidence requirements, and safe-state rules.
- Decision
The runtime returns allow, hold, or block instead of letting action drift through a dashboard.
- Effect
If allowed, the downstream effect is bound to the same governed action record.
- Proof
The packet preserves request, policy context, decision, release path, effect record, and verifier result.
- Verify
A buyer can inspect the record outside the dashboard and decide whether the proof holds.
More governed actions
The same proof loop applies beyond robotics.
Choose a buyer path