# AGENTS.md — TFIS AI Ethics in Education

> **Project:** Course: Navigating Ethical Challenges of AI in Student Learning
> **Course dates:** 1–3 September 2026 (Tuesday–Thursday) · 7 sessions × 2 hours = 14 contact hours
> **Audience:** Higher-education / pre-university educators, academic staff, programme leaders (Malaysian HE)
> **Status:** 🟢 **Content complete** — all 7 slide decks, facilitator guide, assessments, course pack, print pack, and sources are built and ready for delivery.

---

## 1. What This Workspace Contains

The complete curriculum for TFIS's AI Ethics in Education course (workshop), delivered over three days at a Malaysian higher-education institution. The course is a single argument in seven movements: **understand the AI-in-education landscape → build ethical fluency → ship an institutional AI-use guideline.**

### Deliverable tree

```
tfis-ethics-edu/
├── AGENTS.md                             ← this file
├── README.md                             ← project readme
├── course-ai-ethics-course-pack.md      ← standalone master pack (monorepo root, also copied into sources/)
├── reports/
│   └── MIT-AI-Committee-Final-Report.pdf  ← MIT AI in Education report (Aug 2026)
└── TFIS AI Ethics in Education/
    ├── PROJECT-BRIEF.md                  ← project brief (updated through build-out)
    ├── 01-syllabus/
    │   └── syllabus.md                   ← v1.0 — the contract: LOs, commitments, session specs
    ├── 02-slides/
    │   ├── s1.html  (350 lines)          ← The AI Revolution in Higher Education
    │   ├── s2.html  (341 lines)          ← Understanding AI Ethics (case-first: Ofqual + St George's)
    │   ├── s3.html  (314 lines)          ← Academic Integrity in the AI Era (case-first: Texas A&M)
    │   ├── s4.html  (207 lines)          ← Workshop Part 1: Rubric builder + feedback assistant
    │   ├── s5.html  (250 lines)          ← Workshop Part 2: Crutch effect + redesign sprint
    │   ├── s6.html  (241 lines)          ← Institutional Guidelines Part 1: Diagnose
    │   └── s7.html  (229 lines)          ← Institutional Guidelines Part 2: Ship
    ├── 03-facilitator-guide/
    │   ├── facilitator-guide.md          ← Delivery-day companion (run sheets, deck cues, logistics)
    │   └── prompt-handouts.md            ← P01–P15, print-ready per-participant handouts
    ├── 04-assessments/
    │   ├── templates.md                  ← D1–D6 instrument templates + S8 summative rubric + scenario cards
    │   ├── eai-cmm.md                    ← EAI-CMM: Ethical AI Educator Capability Maturity Model
    │   └── capstone-brief.md             ← One-pager: what participants ship (handed out S1)
    ├── 05-print-pack/
    │   ├── print-pack.html (473 lines)   ← A4 print-ready pack (HTML → PDF workflow)
    │   └── print-pack.pdf                ← Generated PDF
    └── sources/
        ├── source-register.md            ← Restricted source register (Stanford HAI primary)
        ├── course-ai-ethics-course-pack.md  ← v1.0 — full course pack (scripts, prompts, appendices)
        ├── AI_Governance_Bill_Fact_Sheet.md  ← NAIO fact sheets (S6–7 grounding)
        ├── malaysia-ai-governance-note.md    ← Crosswalk: Bill ↔ course principles (S6)
        └── cases/
            ├── ofqual-2020.md            ← UK grading algorithm (opens S2)
            ├── st-georges-1980s.md       ← Admissions bias (S2 historical anchor)
            └── texas-am-commerce-2023.md ← ChatGPT false-accusation (opens S3)
```

---

## 2. Arc & Philosophy

| Day | Movement | Sessions | Output |
|-----|----------|----------|--------|
| Tuesday (1/9) | **UNDERSTAND** | S1, S2 | EAI-CMM baseline; ethics triage; case memo |
| Wednesday (2/9) | **BUILD** | S3, S4, S5, S6 | Verification logs; redesigned assessment; guardrailed tutor; gap analysis |
| Thursday (3/9) | **SHIP** | S7 (+S8) | Guideline v0.1; EAI-CMM delta; 90-day action plan |

**Core claims (the seven axioms):**
- **A1:** The gap is the curriculum (capability compounds; governance lags)
- **A2:** The unit of integrity is the assessment, not the student
- **A3:** Literacy precedes governance (you cannot regulate what you cannot operate)
- **A4:** Detection is a losing regime; disclosure + design is the successor
- **A5:** Performance ≠ learning (the crutch effect)
- **A6:** AI allocates attention; humans keep authority
- **A7:** Policy is a product (ship v0.1)

**Two spines run throughout:**
- **Competency:** The 4Ds (Delegation, Description, Discernment, Diligence) from the AI Fluency Framework (CC BY-NC-SA)
- **Maturity:** EAI-CMM — a six-level maturity ladder measured on S1 (baseline) and S7 (retake)

---

## 3. Design Decisions Binding on All Deliverables

1. **Case-first for S2 and S3.** No principle/evidence slide before the room has worked an opening case cold. S2: Ofqual 2020 + St George's 1980s. S3: Texas A&M–Commerce 2023.
2. **The course pack is the spine.** All scripts, prompts, and instruments originate from `sources/course-ai-ethics-course-pack.md`. Syllabus reorders S2/S3 where case-first requires it.
3. **Sources restricted to the register.** Only Stanford HAI (primary), Anthropic, MIT (AI & Education report), Penn/Wharton, Oxford, Harvard — nothing else cited anywhere.
4. **S6–S7 grounded in Malaysian context.** PDPA 2010 as statutory floor, AI Governance Bill fact sheets for risk-tier mapping, `[LOCAL]` slots for MQA/MOHE/senate.
5. **Single-file HTML slides** on the tfis-fluency brand system (cream/pine/sun palette, Archivo + Hanken Grotesk).
6. **Evening rule (S2, S6):** No lecture block >12 minutes. Discussion-heavy, more work-surface slides.
7. **Every live demo ships with offline screenshot fallbacks** in the deck appendix.

---

## 4. The Capstone (Why Everything Exists)

Every participant leaves with:
- **Guideline v0.1** — a ≤2-page draft institutional AI-use guideline with 7 components, version block, red-team backlog, and AI-use disclosure
- **EAI-CMM delta** — measured movement from S1 baseline to S7 retake
- **Personal 7/30/90-day action plan** with named ally, dates, and a kill criterion
- **Portfolio of 6 artifacts** (EAI-CMM, verification log, redesign sheet, guardrailed tutor prompt, guideline, action plan)

---

## 5. File Conventions

| Convention | Rule |
|------------|------|
| **Slide naming** | `02-slides/s{1..7}.html` — one single-file HTML per session |
| **Prompts** | P01–P15, model-agnostic, `[BRACKETS]` for fill-in-values |
| **Instruments** | D1–D7: exit tickets, verification log, redesign sheet, case memo, guideline skeleton, action plan, EAI-CMM score sheet |
| **Session structure** | All specs in `01-syllabus/syllabus.md` — the binding contract |
| **Run sheets** | `03-facilitator-guide/facilitator-guide.md` — operational layer (timings, deck cues) |
| **Course pack** | `sources/course-ai-ethics-course-pack.md` — full scripts, talking points, source register, appendices |
| **Brand** | Cream (`#FAF5EB`) / Pine (`#16453A`) / Sun (`#F2B63C`) — Archivo (headings) + Hanken Grotesk (body) |
| **S2/S6 decks** | Obey the evening rule; physically order case slides before framework slides |
| **Case dossiers** | `sources/cases/` — one per opening case: factual core, discussion prompts, facilitator notes |
| **Print pack** | `05-print-pack/print-pack.html` → PDF — A4 printable, cover page, all instruments, all prompts |

---

## 6. Key Evidence Anchors

| Study | Where | What it proves |
|-------|-------|----------------|
| 2026 AI Index (Stanford HAI) | S1, S3, S6 | Capability/gap widening; 88% org adoption; 4 in 5 students using AI; 6% of policies clear |
| **MIT AI in Education Report (Ad Hoc Committee, Aug 2026)** | **S1, S2, S5, S6** | **8 principles, 30 sub-recommendations; apprenticeship-displacement risk; grading-system provocation; guideline component evidence base** |
| Liang et al., *Patterns* 2023 | S3 (evidentiary core) | 61%+ false-positive bias vs non-native English writers; one-prompt evasion |
| Bastani et al., *PNAS* 2025 | S5 (the keystone) | Crutch-effect RCT: +48%/+127% assisted, −17% unassisted; guardrails engineered harm away |
| Wang & Demszky, Tutor CoPilot 2024 | S1 | Real-time AI-assisted tutoring: +4pp mastery overall, +9pp for lowest-rated tutors |
| Anthropic Education Report, Apr 2025 | S1, S2, S3 | ~50% of student AI use is *direct* answer-seeking; misuse patterns at scale |
| Ofqual 2020 · St George's 1980s · Texas A&M 2023 | S2, S3 opening cases | Automated judgment failures pre-/post-AI; bias laundered into consistency; detection-chaos |

---

## 7. Build Notes for Future Work

- **Case dossiers are compiled** (Ofqual, St George's, Texas A&M) — all three openers are ready.
- **AI Governance Bill grounding is prepared** — fact sheets crosswalked to course principles in `malaysia-ai-governance-note.md`.
- **ALL 7 slide decks are built** as single-file HTML on the tfis-fluency brand system.
- **MIT AI in Education Report** (Ad Hoc Committee, Aug 2026) — PDF available at `reports/MIT-AI-Committee-Final-Report.pdf`; referenced in S1, S2, S5, S6 and cited in the source register under a new MIT section.
- **Course pack is complete** — all 15 prompts (P01–P15), all 7 instruments (D1–D7), full scripts, appendices.
- **Print pack is built** — HTML + generated PDF.
- **No git repository yet** — first init and commit recommended after delivery review.
- **Post-delivery review:** Syllabus states review date 4 Sep 2026 (after 1–3 Sep delivery).
- **Slide count notes:** S1 (350 lines) and S2 (341 lines) are longest — they carry the evidence walk and case-first openings. S4 (207 lines) is leanest per the 20-min-instruction/100-min-doing design constraint.

---

## 8. AI-Assisted Development

This course was drafted with AI assistance (Claude) under human direction and verification. All sources are restricted to the register in `sources/source-register.md` — practicing the diligence the course preaches. The course materials carry their own AI-use disclosure statements throughout.

---

*AGENTS.md v1.0 · 7 Jul 2026 · Owner: Khalil (TFIS)*