Safeguarding Lead — Local AI Usage Guide

Child Protection, Incident Management & Regulatory Compliance

NAS Jiaxing | Updated: 2026-06-19

← All Roles VP Guide → SLT Guide → Compliance Guide →

⚠️ The Danger: How Safeguarding Currently Uses Cloud AI

Real Incident (Names Changed)

A Safeguarding Lead used a cloud AI tool to summarize a complex child protection case. The document contained:

  • Child's full name and date of birth
  • Detailed abuse allegations
  • Family history and social services involvement
  • Medical and psychological reports
  • Police reference numbers

Result: Highly sensitive child protection data processed on US servers. Criminal breach of safeguarding obligations. Potential regulatory action.

Common Cloud AI Misuse by Safeguarding

TaskWhat Gets LeakedRisk Level
Summarizing child protection reportsChild's identity, abuse details, family historyCRITICAL
Drafting referral lettersMedical details, social services info, police dataCRITICAL
Analyzing patternsMultiple children's data, statistical trendsCRITICAL
Training materialsCase studies with real detailsHIGH
ChronologiesTimeline of events with names and datesCRITICAL

✅ The Solution: Local AI for Safeguarding

Why Local AI Changes Everything

  • Zero Data Exposure: Child protection data never leaves the server room
  • Full Audit Trail: Every query logged with your user ID and timestamp
  • Auto-Encryption: All safeguarding queries automatically encrypted
  • Role-Based Access: Only Safeguarding Leads can access CP data
  • PIPL Compliant: Data never leaves China's jurisdiction
  • Court-Admissible: Complete audit logs for legal proceedings

📋 Detailed Usage Scenarios

CHILD PROTECTION

Incident Report Summarization

Current (Dangerous): Upload full CP report to cloud AI

Local AI Method: Use anonymized case numbers and generic descriptions

# PROMPT: CP Report Summary # NEVER include names, dates, or identifying details Summarize this child protection incident for: - Case number: [ANONYMIZED REFERENCE] - Incident type: [general category only] - Actions taken: [general description] - Follow-up required: [yes/no + general type] Remove: names, dates, specific locations, medical details.
📥 Download Template
REFERRAL

Referral Letter Drafting

Current (Dangerous): Use cloud AI to draft referral letters with full details

Local AI Method: Generate template, then add specific details manually

# PROMPT: Referral Letter Template # Generate generic template, then add specific details locally Generate a professional referral letter template for: - Recipient: [social services / police / medical] - Purpose: [initial concern / update / urgent] Include: professional greeting, standard sections, closing.
📥 Download Template
TRAINING

Training Material Development

Current (Risky): Use real case studies in cloud AI

Local AI Method: Generate fictional scenarios with no real data

# PROMPT: Training Scenario # Completely fictional, no connection to real cases Create a safeguarding training scenario for: - Topic: [recognizing signs / reporting procedures / escalation] - Audience: [teachers / support staff / leadership] - Length: [brief / detailed] Use: completely fictional names, dates, and details. Include: scenario, discussion questions, correct actions.
📥 Download Template
CHRONOLOGY

Chronology Organization

Current (Risky): Upload timeline to cloud AI for formatting

Local AI Method: Use anonymized event codes

# PROMPT: Chronology Format # Use event codes, never names or dates Organize this chronology for: - Case reference: [CODE ONLY] - Events: [coded references] - Format: [timeline / table / summary] Remove: names, specific dates, locations, identifying details.
📥 Download Template

🛠️ Open Source Tools & Alternatives

TaskCloud Tool (Risky)Local Alternative (Safe)
Document draftingChatGPT, Claude, GeminiOllama + Qwen/Llama (self-hosted)
Case managementGoogle Docs, NotionNextcloud (self-hosted) + encryption
CommunicationEmail, WhatsApp, WeChatSecure internal messaging (Matrix/Rocket.Chat)
File storageGoogle Drive, OneDriveEncrypted local NAS + Tailscale
Audit loggingThird-party servicesLocal audit system with tamper-proof logs

Our Systems at NAS Jiaxing

We have already built and deployed these tools:

  • HER (Enterprise Server): AMD Ryzen 9, 96GB RAM, RX 7900 XTX — handles all major AI tasks
  • HIM (Standard Server): AMD Ryzen 7, 32GB RAM, 8TB storage — fleet management, backups
  • Grammar Platform: 0604.ai — Custom grammar and vocabulary tool
  • Quiz System: Live at Kahoot-style system
  • Backup Infrastructure: 4-layer architecture (GitHub + GitLab + HIM + Direct Link)

All data stays on school premises. Full audit logs. Zero cloud exposure.

📥 Downloadable Resources

📄 Safeguarding Prompt Library

20+ ready-to-use prompts for CP, referral, and training tasks

Download

📊 Incident Report Template

Template with anonymized fields and AI-ready structure

Download

🔒 Data Handling Protocol

What to remove before using AI — quick reference

Download

🔗 Related Resources

🏗️ Local AI Deployment

Full workshop on building your own AI server

🖥️ AI Fleet Overview

HER, HIM, and all school machines

🛡️ Compliance Guide

PIPL checklist and safeguarding procedures