# Source Register — AI Ethics in Education
Distilled from course pack Appendix B. **Restricted set: nothing outside this register is cited anywhere in the course.** Per-session mapping matches syllabus §4.

## Primary — Stanford HAI

## MIT — AI & Education Report
| Source | Used in |
|--------|---------|
| MIT Ad Hoc Committee on AI in Education, *Final Report* (Aug 2026) — aiandeducation.mit.edu/report/ — 8 principles, 30 sub-recommendations. Full report PDF in `reports/MIT-AI-Committee-Final-Report.pdf` | S1 (recommended reading), S2 (apprenticeship-displacement Card 9), S5 (grading-system provocation), S6 (guideline component 6, instructor disclosure, mirror-test; evidence base for components 4b and 8) |

## Primary — Stanford HAI
| Source | Used in |
|---|---|
| 2026 AI Index Report, esp. Education chapter — hai.stanford.edu/ai-index/2026-ai-index-report | S1 (data walk), S3 (adoption), S6 (policy-clarity stat) |
| HAI Education programs & mission — hai.stanford.edu/education | S1, S2 (human-centered framing) |
| "AI-Detectors Biased Against Non-Native English Writers" (HAI News) | S3 |
| "AI's 'Delusional Spirals'…" (HAI News, Apr 2026) | S2 (wellbeing risk) |

## Stanford research
| Source | Used in |
|---|---|
| Liang, Yuksekgonul, Mao, Wu & Zou (2023), *Patterns* 4(7), DOI 10.1016/j.patter.2023.100779 — detector bias vs non-native writers | **S3 evidentiary core**; cited inside Guideline v0.1 (S7) |
| Wang, Demszky et al. (2024), Tutor CoPilot (arXiv/working paper) | S1 (opportunity ledger) |

## Ivy League / top-tier
| Source | Used in |
|---|---|
| Bastani et al. (2025), *PNAS* 122(26), DOI 10.1073/pnas.2422633122 — crutch-effect RCT | **S5 core**; previewed S2 |
| Mollick & Mollick (2023), "Assigning AI" (SSRN) — seven roles | S5 |
| Kosmyna et al. (MIT Media Lab, 2025), "Your Brain on ChatGPT" (arXiv preprint, N=54 — **always cite with limitations**) | S2 (cognitive risk), S5 (discernment contrast vs RCT) |
| Floridi et al. (2018), "AI4People," *Minds and Machines* — five principles | S2 |
| Wan et al. (2023), reference-letter gender bias, EMNLP Findings | S2 (P03 demo grounding) |
| Harvard CS50 AI tutor (Malan et al., SIGCSE) | S1, S3 |
| Stanford / Harvard / MIT generative-AI teaching guidance pages — devolved-but-scaffolded shape | S6 |

## Frontier lab — Anthropic
| Source | Used in |
|---|---|
| AI Fluency Framework (Dakan, Feller & Anthropic, CC BY-NC-SA 4.0) — aifluencyframework.org | S4 spine; optional pre-work (S1) |
| Anthropic Education Report: How University Students Use Claude (Apr 2025) | S1, S2, S3 (usage + misuse patterns) |
| Anthropic faculty-usage education report (Aug 2025, ~74k conversations) | S4 |

## Local statutory / regulatory (structural, not "cited")
| Source | Used in |
|---|---|
| Personal Data Protection Act 2010 (Malaysia) | S2 (privacy risk), S4 (lab rule), S6–S7 (guideline component 6) |
| NAIO AI Governance Bill consultation fact sheets (`AI_Governance_Bill_Fact_Sheet.md`, this folder; distilled in `malaysia-ai-governance-note.md`) | S6–S7 (five principles crosswalk, three-tier risk framework, Central Authority pattern) |
| `[LOCAL]` slots: MQA programme standards, MOHE guidance, institutional senate | S6–S7, filled by guideline owner |

## Case dossiers to compile (`sources/cases/`)
| Case | Opens | Status |
|---|---|---|
| UK Ofqual 2020 grading algorithm (39% downgraded; U-turn) | S2 (case-first opening) | ☑ `cases/ofqual-2020.md` |
| St George's Hospital Medical School admissions screening (1980s) | S2 (historical anchor) | ☑ `cases/st-georges-1980s.md` |
| Texas A&M–Commerce ChatGPT false-accusation (May 2023) | S3 (case-first opening) | ☑ `cases/texas-am-commerce-2023.md` |

## Framework lineage (disclosed, not cited)
- EAI-CMM adapts the SSA-CMM maturity-ladder pattern (The Future Is Solo).
- 4D competency spine adapts the AI Fluency Framework (CC BY-NC-SA 4.0 — remix rights disclosed to participants in S4).