AI That Works  ·  AI6 Guidance
AI6-T06.1
Per-System Template — Pillar 06
Human Oversight & Intervention Protocol
Document No.AI6-T06.1Version[1.0]
Document Owner[Name / Role]Date[DD/MM/YYYY]
Review Date[DD/MM/YYYY]Status[ Draft / In Review / Approved ]
Organisation[Organisation Name]
Complete one protocol for each deployed AI system. The protocol defines exactly who is responsible for human oversight, under what circumstances intervention is required, and how the system can be overridden or stopped. This document must exist before any system is deployed.
1. System Details
System name and risk classification
[System name]  ·  Risk Classification: [ High / Medium / Low ]
Oversight mode
[ In-the-loop — human must approve before AI output is acted upon ]  |  [ On-the-loop — human monitors and can intervene; AI outputs are acted upon unless challenged ]  |  [ Specify other arrangement ]
2. Oversight Personnel
RoleNamed IndividualOversight ResponsibilityAvailability Requirement
Primary Overseer[Name][Describe what this person reviews and approves][e.g. Business hours — must respond within 4 hours]
Backup Overseer[Name][Describe backup coverage arrangements][e.g. When primary is unavailable]
Escalation — Executive AI Officer[Name]Escalation for systemic issues or override of high-consequence decisionsSame business day
3. Intervention Triggers
Trigger ConditionRequired ActionTimeframe
[e.g. AI output confidence score < [X]%][e.g. Mandatory human review before acting on output][e.g. Before output is used]
[e.g. AI output relates to a safety-critical decision][e.g. Pause AI processing; escalate to overseer][e.g. Immediately]
[e.g. System performance metric breaches threshold][e.g. Suspend AI system; notify IT and Risk & Compliance][e.g. Within 1 hour of detection]
4. Override & Shutdown Procedure
How to override a specific AI output
[Describe the technical steps to reject / override an AI output and document the decision. Include system location and user access requirements.]
How to shut down the system
[Describe the steps to suspend or disable the AI system in full. Specify who has authority to initiate shutdown and what approval is required.]
Protocol approved by
[Name]  ·  [Date]
Executive Accountable Official
[Name]  ·  [Date]