Questions for supervising consequential automated action.

Frameworks establish the obligation. These questions test whether it operates.

Supervisory frameworks increasingly require human oversight, accountability and control over automated and AI-enabled systems. The questions below ask whether those requirements hold at the point where a specific action becomes consequence.

They are drawn from the criteria for Execution Control Infrastructure and published without commercial condition. Questions are numbered for reference. Indicative evidence is illustrative; an institution may demonstrate the same properties in other ways.

Clearance of the action, distinct from permission of the actor.

• Q3. At the Execution Boundary, is a determination made about whether the specific action is cleared to proceed, separate from whether the actor is permitted?
  Indicative evidence: a documented control distinct from identity, authorisation and policy checks.
• Q4. Can the actor seeking execution determine, alter, satisfy, bypass or substitute the conditions of its own clearance?
  Indicative evidence: design and testing records showing that the acting system cannot influence its own clearance.
• Q5. Is the determination controlled by the function or provider that operates or supplies the acting system?
  Indicative evidence: a governance record showing where control of the determining function sits, and its separation from the operating function.

Location of authority at the point of consequence.

• Q8. For a given consequential action, can the institution demonstrate who held authority over it at the moment it proceeded?
  Indicative evidence: the ability to trace a specific action to the accountable authority in effect at execution.
• Q9. Where an action proceeded that should not have, can the institution identify the point at which it could still have been held?
  Indicative evidence: incident reviews that locate the last preventable point.