TFIS · Sharing Session · August 2026
What to do when your university hasn't published an AI policy yet — grounded in the best evidence we have as of 2026. This is not a lecture. This is your operating manual for a world that changed faster than the rulebook.
Movement 1
Let's start with the number that matters most: four in five university students now use AI for coursework. That's not a prediction. That's the 2026 AI Index from Stanford HAI — the most authoritative measurement we have.
And yet only about half of schools have any AI policy at all. Of those, just 6% of teachers say the policy is clear. This is not a moral failing of your institution. It's what happens when a technology moves faster than governance.
This means you're navigating a space with no official map. The good news: enough research now exists — from Stanford, MIT, Wharton, Anthropic — to build a working map together. That's what this session is.
You are not alone in being uncertain. Every student in this room — every student in the world — is working this out in real time. The question isn't whether you use AI. The question is whether you're using it on your terms or on its.
Movement 2
When no policy exists, the ethical default is not "anything goes." It's the set of principles that every framework — from Stanford's to Malaysia's proposed AI Bill — converges on. These aren't abstract. They're the operating system under every good AI policy.
AI should help you learn, not just produce. The test: does using AI make you better without it?
First, do no harm. AI shouldn't mislead you, fabricate sources, or amplify bias in ways that hurt your work or others.
You stay in charge. Dependence on AI is autonomy decay on an instalment plan — every time you let it decide for you, you rent out a piece of your judgment.
Fair distribution of benefit and harm. Not everyone has access to the same tools, gets flagged by detectors at the same rate, or has their work judged by the same standard.
If you can't explain how AI touched your work, you can't defend it. Transparency is not confession — it's professional practice. In an AI-native workforce, the ability to describe your process is what separates a professional from a prompter.
Movement 3
If your university runs student work through an AI detector — and many do — you need to know what the research actually says about those tools. The short version: they don't work the way most people think they do.
Here's the deeper problem: AI-generated text detection is structurally flawed. The arms race between generation and detection fundamentally favours generation — every improvement to detection prompts a trivial evasion. The Stanford researchers themselves recommend: avoid these detectors in evaluative settings.
Instead of a cat-and-mouse game you can't win, the evidence-based approach is disclosure. An AI-use statement on every submission: what tool, what for, what you verified yourself. This is not a confession — it's professional practice. Anthropic, a frontier AI lab, models this with an "AI Diligence Statement" on their own course materials. If a frontier lab discloses, your university can. So can you.
Does any course you're taking this semester use an AI detector?
Movement 4
This is the most important study you'll see today. It's the one piece of research that should shape every decision you make about AI in your learning.
The lesson is not "AI is bad for learning." It's that design determines outcome. The exact same model (GPT-4) either harmed or didn't, depending entirely on how it was configured. Your job as a learner is to understand this line — and stay on the right side of it.
AI can play different roles (Mollick & Mollick, Wharton). Some threaten your learning; others protect it:
| Role | What it does | Learning risk |
|---|---|---|
| Tool | You produce output through it | ⚠ High — crutch risk |
| Tutor | Guides you, doesn't answer | Medium — depends on guardrails |
| Mentor | Gives feedback on your work | Low — you produce first |
| Coach | Asks metacognitive questions | Very low — increases thinking |
| Student | You teach the AI | Very low — strengthens understanding |
| Simulator | Creates practice scenarios | Low — you act, AI reacts |
| Teammate | Collaborates on a shared task | Medium — depends on your contribution |
In your most common AI use, which role does it play for you?
Movement 5
When a course has no AI rule — which is most courses — what applies? The honest answer is: the default rule. And choosing the right default matters more than any ideal rule.
Every well-designed course declares a lane per assessment. When no lane is declared, the best default is the one adopted by leading universities (Stanford, Harvard, MIT): 🟡 permitted with disclosure. Here's what each lane means for you:
Even when no guideline is published, the following floor exists — by law (PDPA 2010) or by best practice (MIT AI in Education Report, Aug 2026; Stanford HAI):
Your personal data and student work cannot be fed into unvetted third-party AI tools without institutional agreements. If a lecturer asks you to upload your assignment to a free AI tool, ask: does the university have a data agreement with this provider? If not, your PDPA rights may be at stake.
The same disclosure standard should apply to students and staff. If your lecturer uses AI to write feedback on your work, they should disclose it — just as they ask you to. A policy that only regulates one side is not a policy; it's a double standard.
A detector score alone is not sufficient evidence of misconduct. If accused, you have the right to present process evidence — drafts, notes, edit history, your own disclosure — and have that evidence treated as primary. The Stanford Patterns study (61% false-positive rate) should be part of every institution's integrity procedure.
Every AI guideline should have a named owner, a version number, and a review date ≤12 months. If you're told "there's a policy somewhere," ask: who owns it, and when was it last updated? A policy without a maintainer is a wish.
When a course doesn't specify its AI rule, the honest default is: 🟡 permitted, with disclosure. The common but fictional default — "silence means banned" — produces widespread silent violation and erodes trust. Silence means you disclose. That's the August 2026 best practice.
Movement 6
Knowing that AI changes learning isn't enough. You need a system for deciding how to use it in each situation. The 4Ds (from the AI Fluency Framework, CC BY-NC-SA) give you that system:
Decide what to hand off. Not every task should be delegated to AI. Sort by: (a) is this a skill I need to own? (🔴 don't delegate); (b) is this a task I can do but AI can do faster? (🟡 delegate with oversight); (c) is the point of the task to learn AI fluency itself? (🟢 delegate fully).
Set it up well. Garbage in, garbage out — but also: vague in, generic out. Give the model context, constraints, your audience, your voice. A well-described task yields output you can actually use; a poorly-described one costs more time to fix than doing it yourself.
Judge the output. AI is confident even when wrong. Check every factual claim. Look for hallucinated sources. Notice tone mismatches. Flag bias (gender, cultural, language). Discernment is the skill that separates a professional from a prompter — and it only develops through practice.
Document and improve. Keep a record: what did you ask, what did you get, what did you change? A simple verification log — tool used, prompt, output, your edits — is your professional paper trail. Over time, it becomes your personal playbook for using AI effectively.
Here's a template you can use for every submission, adapted from the course's P05 instrument. Fill in the blanks:
Tool(s) used: [name of AI tool(s)]
What I used them for: [brainstorming / drafting / editing / checking / explaining / other]
What I verified myself: [facts / sources / logic / tone / all of the above]
What I kept in my own head: [the argument structure / the creative direction / the core analysis]
Honesty declaration: This submission represents my own understanding. AI assisted where stated; I did not use it to evade learning.
Movement 7
Everything so far is about surviving your current education. But you're not just a student — you're preparing to enter an AI-native workforce. Here's how the frameworks in this session translate to professional life.
The three-lane framework isn't just for courses. It's a lifelong skill:
The ability to explain how AI touched your work — honestly, specifically, without jargon — is becoming the most valued skill in knowledge work. It signals that you:
In a world where anyone can prompt, the person who can audit, critique, and improve the AI's output is the one who keeps the job.
✅ Capstone
You've covered the evidence, the principles, the traps, and the tools. Now it's your turn. This is a short guided plan — your own rules for navigating AI as a learner and future professional. Everything you enter stays on your device.
Customise this for your own use. You'll paste or adapt it for every submission.
These are the skills you're building that no AI should do for you. Check what matters:
A kill criterion is your early warning system. What would tell you your AI use is hurting your learning — not helping?