AI policy
Effective / published version: 2026-09-19
_Last updated: 22 April 2026_ 1. PURPOSE This policy explains how AI is used within LearnSys EduFlow, the controls available to schools, and our responsible AI principles. 2. OUR PRINCIPLES a) Teacher in control — Teachers decide when AI runs, review all output, and remain the final authority on assessment and feedback. b) Student assistance, not substitution — Student-facing AI provides hints, explanations, and scaffolding, not direct answers. c) Transparency — Every AI session is logged with model, tokens used, and timestamp, viewable by administrators. d) Data minimisation — We send only the prompt content needed for the requested task. We do not permit upstream providers to retain data for model training. e) Human oversight — Workflow outcomes that affect students (e.g. AI-suggested marks) require teacher approval unless explicitly disabled by school policy. 3. AI FEATURES IN EDUFLOW Teacher tools: - Lesson Plan Generator - Worksheet Maker - Question Generator - Quiz / Exam Generator - AI Marking Assistant (suggests scores and feedback; teacher reviews before releasing) Shared tools: - AI Chat Assistant (context-aware; respects tenant policy) - Document Vault with semantic search (optional, if RAG enabled) Student tools: - AI Chat in Academic Integrity Mode (hints, not answers; locked during assessments) 4. SCHOOL CONTROLS School administrators can configure the following via Policies (政策設定): - Enable/disable student AI chat. - Enable/disable parent AI chat. - Set Academic Integrity mode (hints only) vs Open mode. - Assessment Lock: disable AI access while students are taking locked assessments. - Allowed apps: restrict which AI tools are available to each role. - Chat retention period: default 90 days, configurable. 5. UPSTREAM AI PROVIDERS We use commercial and self-hosted AI providers for inference. When a prompt is sent to a provider: - We do not transmit student names or identifiers unless necessary for the task. - We require providers to refrain from using our data for model training. - We may switch providers for reliability, cost, or quality reasons. 6. ACCURACY AND LIMITATIONS - AI may produce incorrect, outdated, or biased output. Teachers must review before use. - AI-generated writing detection is not currently offered; teachers should use professional judgement. - AI translations and summaries are not a substitute for qualified interpreters. 7. ASSESSMENT INTEGRITY - Student AI is blocked during assessments flagged with Assessment Lock. - Attempts to access AI during locked assessments are logged for school review. - AI-generated marks are labelled as "AI suggestion" until a teacher approves them. 8. DATA SENT TO AI Typically includes: - The teacher/student prompt text. - Relevant educational context (class year level, subject). - For RAG: selected document chunks from your Vault. Never includes: - Passwords, credentials, or billing details. - Full user lists or unrelated personal data. 9. AUDITING All AI sessions are recorded and retained per the Data Retention Policy. Administrators can export audit logs via EDB Export. 10. CONTACT AI policy questions: [email protected] Report AI concerns: https://hk.learnsys.org/contact-support