Back to Guides

A short AI-narrated film with English captions. Pressing play loads the YouTube player. Watch on YouTube

Guide Origin: Official | ArcKit Version: [VERSION]

/arckit:jsp-936 provides MOD Dependable AI (JSP 936) documentation for defence AI systems. Use it to evidence the five ethical principles, lifecycle coverage, and approval pathway.

Before You Run

Artefact Purpose
Requirements & architecture Identify AI components, data flows, and human interaction
Risk register & security assessment Seed ethical and safety risks
Data model & DPIA Map personal/special category data
Governance roles Confirm RAISO, Ethics Manager, project lead

Command

/arckit:jsp-936 Generate JSP 936 assurance for <AI system>

Output: projects/<id>/ARC-<id>-JSP936-v1.0.md

Assessment Structure

Section Focus Key Evidence
Risk Classification Likelihood × impact (Critical → Minor) Score triggers approval path (2PUS, Defence-level, TLB)
Ethical Principles Human-centricity, Responsibility, Understanding, Bias & Harm, Reliability Actions and mitigations per principle
Lifecycle Phases Planning → Quality Assurance (8 phases) Evidence across planning, build, testing, deployment, monitoring
Governance Roles, sign-off, battle rhythm RAISO, Ethics Manager, assurance board dates
Continuous Monitoring Drift detection, retraining policy, incident response Links to /arckit:servicenow and ops runbooks

Review Playbook

  1. Validate risk classification and approval route with programme governance.
  2. Confirm human-in-the-loop controls and escalation paths.
  3. Check bias and harm mitigations include test coverage and ethical sign-off.
  4. Ensure supplier assurance (if third-party AI) is documented.
  5. Log required actions in backlog with owners and dates.

Update Cadence

Trigger Action
New AI capability or model retraining Re-run /arckit:jsp-936 and update evidence
Risk classification change Notify RAISO/Ethics board; revisit approval pathway
Annual assurance review Refresh assessment; confirm monitoring data

Store signed copies and decisions in the MOD assurance repository for audit.