Learn · Education

The Integrity Question: AI, Rules, and Detectors

The rules are inconsistent, the detectors are unreliable, and the stakes are your record. Here is how to navigate all three.

Academic integrity used to have settled edges: copying was cheating, help was allowed, and everyone roughly knew the difference. AI dissolved those edges in about a year, and institutions are still redrawing them — which means students currently navigate a landscape where the same act is encouraged in one classroom, forbidden in the next, and undetectable-but-career-ending in a third. This guide is about navigating that honestly and safely. It pairs with [studying with AI](/school/studying-with-ai/), which covers the deeper question of not cheating yourself; this one covers the institutional layer: rules, detectors, and self-protection.

Rule one: the policy is per-classroom, and asking is free

The single most practical fact: there is usually no single 'school policy' that settles your situation. Institutions publish frameworks; individual instructors set the binding rules, and they genuinely differ — the writing professor who forbids AI contact with drafts and the one who requires an AI-critique exercise may share a hallway. So the discipline is unglamorous: read each syllabus, and where it is silent or vague — which is still common — ask, in writing, before the assignment. 'Is it acceptable to use AI for brainstorming and outline feedback if the writing is my own?' costs one email. The same question asked *after* an accusation is a defense; asked before, it is a shield. Never extrapolate one instructor's permission to another's classroom.

Rule two: understand what detectors actually are

AI-writing detectors deserve a clear-eyed paragraph, because both students and institutions widely misunderstand them. They are statistical pattern-matchers producing probability guesses, not evidence-grade tests, and their documented failure modes cut in both directions: they can be evaded, and — the part that matters for honest students — they produce false positives, flagging genuinely human writing as AI-generated. The documented pattern is troubling: formulaic prose, careful conventional writing, and especially the writing of non-native English speakers trigger flags disproportionately. Major AI companies' own attempts at detection tools have been withdrawn over accuracy; the serious academic-integrity conversation has been moving away from detector-as-verdict toward process-based assessment for exactly these reasons. What this means for you is *not* 'detectors are fake, relax' — institutions still use them, and a flag still starts a painful process. It means the flag is an accusation you can be hit with while innocent, which makes self-protection rational for everyone:

Write where history accrues. Cloud documents with revision history are a timestamped record of your writing unfolding — the single strongest evidence a wrongly-flagged student can produce.
Keep the debris. Outlines, notes, marked-up sources, earlier drafts. Real work leaves a trail; keep yours until grades are final.
Be ready to defend orally. A student who can discuss their argument, sources, and choices in conversation ends most disputes quickly. (Notice this is also just a description of having actually done the work.)

Rule three: the lines themselves, honestly drawn

Policies differ, but the underlying logic is consistent enough to internalize. Broadly defensible in most contexts: AI as [tutor and study partner](/school/studying-with-ai/), brainstorming and idea development, feedback on drafts you wrote, grammar-level polish, and research assistance whose outputs you verify and cite properly. Broadly indefensible everywhere: submitting generated work as your own writing, AI-completed problem sets in courses building exactly those skills, and fabricated citations — which deserve special mention because AI invents plausible-looking sources fluently, and submitting an unchecked fabricated citation is a genuine, checkable integrity violation even where AI use itself was permitted. The contested middle — AI-assisted drafting with heavy revision, translation of your own work, structural editing — is precisely where classroom policies differ and where the ask-first rule earns its keep.

The disclosure move

A quiet norm worth adopting ahead of requirements: where AI use is permitted, a brief methods note — 'AI was used for outline feedback and citation formatting; all text and analysis are my own' — converts ambiguity into transparency. It reads as integrity rather than confession, it preempts the accusation conversation entirely, and it is exactly the habit [professional and creative work](/create/ai-creative-rights/) is converging on anyway. The students in the strongest position through this whole unsettled era are the ones with nothing to reconstruct: clear permission, real work, visible process, stated methods. That position is available to anyone willing to send one email and keep their drafts.

Common questions

Is using AI for schoolwork considered cheating?

It depends on the specific classroom — instructors set binding rules within loose institutional frameworks, and they genuinely differ. Tutoring, brainstorming, and feedback on your own writing are broadly accepted; submitting generated work as your own and fabricated citations are violations essentially everywhere. Where a syllabus is vague, ask in writing before the assignment.

How accurate are AI writing detectors?

They are probability estimators, not evidence-grade tests, with documented failures in both directions — including false positives on genuinely human writing, disproportionately affecting careful conventional prose and non-native English speakers. Even major AI companies have withdrawn detection tools over accuracy. Institutions still use them, so a flag is a risk even for honest students.

How do I protect myself from a false AI accusation?

Build the paper trail as you work: write in cloud documents that keep revision history, retain outlines, notes, and earlier drafts until grades are final, and be prepared to discuss your argument and choices in conversation. A visible process ends most disputes quickly — and doubles as simply having done the work.

Should I disclose AI use on assignments?

Where AI is permitted, a brief methods note — what AI did, what remained yours — is increasingly the strong move: it converts ambiguity into transparency, preempts accusation conversations, and matches the disclosure norms professional work is adopting. Where AI is prohibited, the disclosure question is moot; the use itself is the violation.