AI and academic integrity in 2026: what you can and cannot do

AI tools have changed how most students study in two years, and the schools and exam authorities have responded patchily. This guide takes the question head on. Where is the line? What does each state allow? And what should you actually use AI for to learn better, not to learn less?

The bright lines first

Three things are clearly cheating in 2026 in every Australian curriculum. Each is treated as malpractice or a breach of assessment rules, with penalties set by the authority and your school:

1. Submitting AI-generated work as your own. Whether the AI wrote it or you rewrote what the AI wrote, if substantive ideas, structure, or prose came from a model and you presented it as yours on an assessed task, that is plagiarism. The fact that AI does not show up in a Turnitin originality check does not change what it is.
2. Using AI during an exam where it is not permitted. This includes phones in your pocket, smart watches with assistants, hidden earpieces. Every state exam treats this as serious misconduct.
3. Helping someone else cheat with AI. Generating answers and sending them to a friend in a different exam slot, sharing prompts that produced graded work. Both students get penalised.

These are not new rules. They are the same rules that applied to copying off a friend or buying an essay from an essay mill, with AI added as the new mechanism. The mechanism changed; the principle did not.

What each authority actually says

The detail differs by state and is changing quickly. As of 2026:

NESA (NSW HSC)

HSC exam rules don't allow electronic devices other than approved calculators. For school-based assessment, NESA says marks must reflect the student's own knowledge, skills and understanding and must not reflect a student's use of generative AI. Because of AI, NESA has limited take-home assessment: school-based assessment programs can include no more than one take-home task, worth at most 15%, for HSC courses from Term 4 2026 and Preliminary courses from Term 1 2027. NESA also tells schools not to rely on AI-detection tools as the main safeguard against malpractice.

VCAA (Victoria VCE)

Unattributed or disallowed use of generative AI in assessment may be a breach of academic integrity and is investigated under your school's policy for authenticating school-based assessment. From 2026 the Extended Investigation declaration includes a statement about generative AI use.

QCAA (Queensland QCE)

Work submitted must be your own, and unacknowledged AI use can be academic misconduct. From 2026, Year 12 graduates must also have completed an academic integrity course (such as QCAA's online course in the myQCE Student Portal) to be eligible for a QCE. External assessments are sat under exam conditions without devices.

The unifying principle: the work counts because it represents what you can actually do. If the work was produced by another tool or person, the credential it counts toward is misleading. The institutions, sensibly, take this seriously.

If you are ever unsure whether a specific tool is allowed for a specific task, ask the teacher who set it. In writing. Their answer is your protection.

What AI is actually good at

Used well, AI is a study accelerator. The places where it produces real learning gains:

Asking it to explain something you do not understand

"I have read this paragraph from my chemistry textbook and I do not get why X. Walk me through it slowly with a concrete example." This is the closest thing most students have to a private tutor. The model has flaws (more on that below), but the back-and-forth conversation is genuinely useful, especially at 11pm when no tutor is awake.

Quizzing yourself

"Quiz me on Module B Hamlet themes. Ask me a question. I will answer. Then mark me and tell me what I missed." This is active recall on demand. It is dramatically more useful than re-reading your notes.

Generating practice problems

"Make me 10 practice questions on the chain rule at the level of a hard HSC Mathematics Advanced question." Then you do them, closed book, and check your answers. Practice volume is one of the biggest predictors of maths performance, and AI lowers the cost of generating practice.

Outlining your thinking before you write

Not "write me an essay on Hamlet's delay." That is the line. But "I am going to write an essay arguing that Hamlet's delay reflects an interrogation of certainty, and I want to talk through whether my structure makes sense, what the counter-arguments are, and what evidence I should be looking for in the text." That is using the AI the way you would use a study partner who happens to know the play.

Translating a confusing question

"I do not understand what this exam question is asking. Can you rephrase it in three different ways?" Reading comprehension on exam questions is itself a skill, and AI can help you train it.

Feedback on your own writing

Once you have a draft, "Read this and tell me which paragraphs are weakest and why" is genuinely useful. Critically: you wrote the draft. The AI is reacting to your work, not producing it.

What AI is bad at

The same tools also have real failure modes that students underestimate.

Hallucination

AI tools sometimes invent facts, quotes, dates, citations. They do this with the same confident tone they use for accurate information. For HSC History, this is especially dangerous: dates, statistics, and quotes that look plausible are often subtly wrong or completely fabricated. Always verify with the textbook or syllabus document.

Australian context blind spots

Many AI tools were trained largely on US/UK content. They will sometimes hallucinate that the ATAR works like the US SAT, or that "module" in HSC English means what "module" means in a US course. If the AI tells you something about how the HSC, VCE, or QCE works and it sounds slightly off, it probably is.

Subject-specific syllabus mismatch

AI does not know your school's exact reading list or your teacher's specific marking rubric. It knows the general shape of HSC English Module B; it does not know your school's interpretation of it.

Producing the prose for you

This is the failure mode that matters most for your learning. If you let the AI write your sentences, you do not build the muscle of writing those sentences. Trial week and the final exam happen with a pen in your hand and no AI. The students who used AI to write their drafts in August find themselves staring at a blank page in November, having spent three months not training the actual skill the exam tests.

If the AI's role in your study is "I tell it the topic and it gives me the writing," you have a learning problem dressed up as efficiency. The exam will tell on you. Use the AI to think with, not to write for you.

A working rule of thumb

Before you use AI on a piece of school work, ask yourself: what skill is this task supposed to be measuring, and would using AI here let me skip practising that skill?

For a homework reading comprehension: the skill is reading and understanding. Using AI to summarise the reading skips the skill. Don't.

For a maths problem set: the skill is solving the problems. Using AI to solve them skips the skill. Do them yourself, then ask AI to explain anything that went wrong.

For an essay assessment: the skill is constructing an argument with textual evidence in your own voice. Using AI to draft paragraphs skips the skill. Use AI to talk through your thesis, to challenge your structure, to flag weak paragraphs in your own draft. Don't ask it to write.

For an exam revision quiz: the skill is recall under pressure. Using AI to generate practice questions trains exactly the skill. Go for it.

The pattern: AI is great for scaffolding around the practice, and dangerous when it replaces the practice. The exam is not testing whether you can summon a model. It is testing what you can do under pressure with nothing but a pen.

What to do if you have already used AI badly

If you have submitted a piece of assessed work that was substantially written by AI, the calculus changes depending on whether you have been caught yet.

Before submission, before being caught

Withdraw the submission if you can. Talk to the teacher honestly. Most schools will let you redo the task, sometimes with a late penalty, far less seriously than they would treat actual misconduct. The honest conversation now is a much better outcome than getting caught later.

After submission, before being caught

Do not panic, but understand the risk. Most schools and authorities can detect AI-written work in ways that are not always visible to students (writing style mismatch, perplexity scores, follow-up oral questioning, comparison with previous work). The longer it sits, the worse the eventual conversation if it comes. If your conscience is weighing on you, talk to a teacher you trust.

After being caught

The penalty depends on the assessment, the state, and the school. Cooperate honestly with the investigation. Lying about it tends to make the consequence dramatically worse than admitting and accepting. School counsellors and wellbeing staff can support you through the process; this is not the end of the world even if it is treated seriously.

A note on AI-detection tools

Some schools use AI-detection tools, but they can produce false positives. NESA's 2026 guidance says detection software should be only one input to a teacher's professional review, not the sole proof of AI use. Schools also use:

• Comparison with your previous in-class writing.
• Oral questioning about your submitted work.
• Process artifacts (drafts, version history, in-class writing on related topics).

The combination is harder to game than detection tools alone, and authorities are redesigning assessment (for example NESA's limit on take-home tasks) so that more of it is done under supervision.

Where ExamExplained sits

ExamExplained offers AI-assisted study tools, so the same rules apply to them: use them to explain, quiz and give feedback on your thinking, and write the assessed work yourself.