TFIS · AI Ethics in Education · Course
Academic Integrity
in the AI Era
Managing Plagiarism, Bias and Copyright Issues
Session 3 · Wednesday 2 September 2026 · 8.30–10.30 am
Opening · 10 min
Your words first
- Three case memos, read aloud by their authors.
- Then: answers to the three most common exit-ticket questions from yesterday.
A true story · no verdicts yet
Texas,
May 2023
Graduation week at Texas A&M University–Commerce. An instructor suspects his class used ChatGPT.
The case · the method
His detector: ChatGPT itself.
- He pasted student essays into ChatGPT and asked whether it had written them.
- ChatGPT — which has no capability to recognise its own past output — obligingly "confirmed" authorship. Essay after essay.
- On that basis: submissions failed, and graduating seniors had diplomas temporarily withheld while the university investigated.
Case dossier: sources/cases/texas-am-commerce-2023.md
The case · how it ended
- Students protested innocence. One produced timestamped drafts — and had the only defensible evidence in the room.
- The university stated no students ultimately failed or were barred from graduating — and that it was developing AI policies. At the time of the incident, it had none.
- The instructor wasn't a villain. He was improvising in a policy vacuum — so his improvisation became the policy, for an entire class, at the worst possible moment.
Work the case · tables · 10 min
Before any data:
- List every point where this could have been stopped. Which was cheapest?
- Would this survive your institution's appeals process? Is the detector's evidentiary status written down anywhere?
- A student in that class honestly never used AI. Describe their week.
Reframe
SAY
"Plagiarism rules were never the point. They were a proxy — a cheap test for an expensive
question: did learning happen inside this student? For seventy years the proxy held because
producing text was hard. AI made text free, and the proxy snapped. You now have two options: rebuild
the proxy with detection technology, or go after the real question directly with assessment design.
This morning I'll show you why option one is a trap — with numbers."
The detector evidence · deliver slowly · 1 of 3
The bias result
61%+
of 91 real TOEFL essays by non-native English speakers falsely flagged as AI-generated — on average, across seven widely used detectors
~0
false-positive problem on essays by US eighth-graders. Near-perfect on native-speaker text.
Liang, Yuksekgonul, Mao, Wu & Zou, Patterns 4(7), 2023 — Stanford
The detector evidence · 2 of 3
The mechanism: it isn't detecting AI
- Detectors lean on perplexity — roughly, how predictable the word choices are.
- Non-native academic writers naturally use more constrained vocabulary and syntax.
- The detector is detecting limited lexical variety — which is to say, it's detecting your international students and your ESL students.
In a Malaysian university, where most students write English as a second or third language, this is not an edge case. It is the main case.
The detector evidence · 3 of 3 · same paper
One prompt flips the verdict
"Enhance the word choices to sound more like a native speaker" — misclassified essays flipped back to "human."
- The detector punishes honest non-native writers and passes dishonest users who add one laundering step.
- Biased in one direction. Evadable in the other. The worst possible combination for an evidentiary instrument.
- The researchers' own recommendation: avoid these detectors in evaluative settings. The arms race structurally favours generation.
Liang et al., Patterns 2023; companion Patterns commentary
Before anyone relaxes
The misuse is still real
Do not swing to denial: Anthropic's Education Report documents students requesting test answers and detector-evading rewrites at scale.
The problem is genuine. The tool is wrong.
Anthropic Education Report, Apr 2025
Exercise · Detector on Trial · 30 min
The motion
"This institution should treat AI-detector scores as admissible primary evidence in integrity proceedings."
- Each table splits: two argue for, two against, one judges.
- Twist: the for side goes to those who voiced anti-detector views — and vice versa. Steelmanning is the point.
- 8 min prep · 4+4 min arguments · 2+2 rebuttal · judges rule with a one-sentence ratio.
Debrief · facilitator
Expected convergence
Detector output is at most a screening signal that triggers human process — never proof.
If a table rules otherwise, one question for the judge:
ASK
"Which of your own students is most likely to be falsely flagged?"
Let the silence do the teaching.
If not detection, then what?
The successor regime:
process evidence
Integrity signals produced during creation — not inferred after.
Five mechanisms
| Mechanism | Why it works |
|---|
| 1 · Version trails | Drafts, document history, commit logs. Effort leaves fingerprints; laundering doesn't. |
| 2 · Oral defense sampling | 5-minute vivas for a random 20% of submissions. Students who did the work pass easily; deterrence generalises to 100%. (P06 generates the questions.) |
| 3 · In-class anchors | Some fraction of every assessment executed live. S5 formalises this as the three-lane design. |
| 4 · Disclosure as norm | An AI-use statement on every submission — what tool, what for, what was verified. Not a confession; a norm. Anthropic's own courses ship an "AI Diligence Statement." If a frontier lab discloses, your students can. So can you. |
| 5 · The design dividend | Every hour moved from policing artifacts to designing process buys better evidence and better pedagogy. Detection buys neither. |
Live build · P05 · 10 min
Draft your disclosure template — now
Run P05: a 4-line AI-use disclosure template for student submissions in your course — tool(s) used, what for, what the student verified themselves, one-line honesty declaration. Plain language, no legalese.
Then the parallel version for staff use on teaching materials. Symmetry is credibility.
This template goes into your team's guideline on Wednesday — component 4, near-verbatim.
Copyright fault lines · the map, not legal advice
- Training-data provenance — contested in ongoing litigation worldwide. Status unstable: institutional policies should reference principles, not case outcomes that may flip.
- Ownership of AI-assisted output — varies by jurisdiction; substantial human authorship is generally the anchor. Practical rule for teaching materials: the more you transform, the safer you stand — and disclose regardless.
- Student IP — submitting student work into external AI tools without consent raises both copyright and PDPA questions. Institutional accounts + anonymization is the floor. Session 6 encodes it.
The misuse conversation · script it — this protects both parties
SAY
"Walk me through how you made this. Show me your process — drafts, notes, history.
Explain this paragraph's argument in your own words."
- Notice what's absent: no accusation, no detector percentage, no trap.
- A student who did the work demonstrates it in ninety seconds.
- A student who didn't reveals it just as fast — and you now hold process evidence a committee can stand on.
Ticket · one line · handed in
Pre-commit this afternoon's raw material
Name the assessment you will redesign in Session 5 — and which lane pattern you suspect it needs.
Contrarian close
SAY
"The contrarian position, stated plainly: banning AI is the least safe policy available to you.
A ban doesn't stop usage — four in five students are already there. It stops disclosure. It converts
your most honest students into your most disadvantaged ones and hands the advantage to the
laundering-literate. Every ringgit spent on detection is a ringgit spent making adversaries of your
students. This afternoon we stop policing and start building. Minum dulu."
Appendix · facilitator only
Offline fallbacks & notes
- [ insert screenshot: P05 disclosure-template draft run ]
- [ insert screenshot: P06 viva-question generation on a sample excerpt ]
- Do not name the Texas A&M instructor on slides — the lesson is systemic, not personal.
- Keep case claims conservative: pasted essays into ChatGPT and asked; diplomas temporarily held; university said no one ultimately failed or was blocked.