AI Safety Pack (PeopleSafetyLab)

Version: v1.0 (2026-02-24)

Goal: a lightweight, client‑ready starter kit to deploy AI responsibly (people‑harm + data leakage focus) in 30 days.

Contents

1. 00-quick-start-guide.md — START HERE: 48‑hour emergency deployment guide
2. 01-ai-use-policy.md — AI use policy (employee/contractor)
3. 02-approved-prohibited-usecases.md — executive 1‑pager (allowed vs not)
4. 02a-ai-use-case-matrix.md — intake/approval matrix + evidence artifacts
5. 02b-decision-tree.md — visual decision tree for rapid classification
6. 03-people-harm-risk-register.md — risk register template (people‑harm lens)
7. 04-controls-map.md — mapping risks → controls (owners + evidence)
8. 04a-controls-quick-reference.md — one‑page control lookup (owners + evidence + bundles)
9. 05-30-day-implementation-checklist.md — rollout plan + evidence checklist
10. 06-training-deck-outline.md — runnable 60–90 min training outline
11. 06a-training-facilitator-guide.md — practical tips for delivering training sessions
12. 07-use-case-register-template.md — lightweight register template
13. 07-use-case-card-template.md — one‑pager per Conditional use‑case
14. 08-exception-decision-record-template.md — time‑boxed exception approval record
15. 09-jurisdiction-notes.md — guidance for EU, US, UK, KSA/GCC, Singapore, Australia
16. 10-qa-scenarios.md — end‑to‑end self‑test scenarios
17. 11-qa-results-scenario-a-b.md — QA results (Scenario A + B)
18. 12-qa-results-scenario-c.md — QA results (Scenario C)
19. 13-communication-templates.md — rollout announcements, manager briefings, training invites
20. use-cases/ — example filled use‑case cards

How to ship this to a client (recommended)

• Step 1 (1–2h): Tailor approved tools + data classifications + incident owner/channel.
• Step 2 (2–3h): Run a 60–90 min workshop: top use‑cases + classify via matrix + draft controls.
• Step 3 (same day): Deliver v1 policy + use‑case register/cards + risk register + 30‑day plan.

Status

• v1.0: consistency sweep complete (control IDs + references + templates + QA docs).
• Tone is intended to be jurisdiction‑neutral (policy/controls guidance, not legal advice).