This policy is written to faculty.
It is not a set of rules handed to students. It is a charge given to those who teach them.
A policy aimed at students is enforcement — it asks what students may not do, and it ends there. A policy aimed at faculty is formation. The teacher carries this into every class, every assignment, and every conversation about students' work. What students receive is not this document but the teaching it produces.
Didactic, not limiting.
Its purpose is to teach. It is not a fence.
A policy that only restricts leaves students with a list of prohibitions and no ability. It also fails on its own terms. The tool is on every phone, so a ban cannot be enforced, and students will use it whether or not we permit it. What an unenforceable rule teaches is not restraint — it teaches that the school's rules are announced but not meant, and students go on using artificial intelligence with no one having shown them how. We lose twice over: the rule is ignored, and the teaching never happens. So we are not asking how to keep artificial intelligence out of the classroom. We are asking how to teach students to use it well.
Where this policy sets a limit, the limit will come with the reason for it, so that students who leave us keep the judgment rather than only the restriction.
A policy about the machine turns out to be about teaching.
Write this policy honestly and it will not stay a policy about a tool. It puts a much larger question to the school: what is a teacher's hour actually for?
For as long as any of us have taught, part of the answer was delivery. The teacher held knowledge the students could not easily reach, and the lesson was where it was handed across. That is finished. Knowledge is now two seconds away, free, and better organised than any lecture. A teacher who still spends the hour transmitting it is spending the scarcest thing in the school on the one thing that has become cheap.
So lecturing is not the way. Explaining is. Guiding is. Coaching is. The teacher stops being the source of the material and becomes the one who shows how the pieces bear on each other, who sets the problem that forces a student to work something out — and who can tell the difference between a student who has understood and a student who has assembled.
And this is where the time comes from. Every practical objection to the way this guide asks teachers to work — that you cannot listen to each student explain their work, that there is no room in the week for it — assumes the hour is already full of delivering knowledge. It does not have to be. The hour freed from delivery is the hour available to sit beside a student and find out what they actually hold.
That is not a burden this policy adds. It is the thing the lecture never left room for. For the first time, knowing our students individually can be the ordinary shape of teaching rather than a kindness squeezed into the margins — we can mentor, and we can actually know whether a student is ready to go further. That is what artificial intelligence has handed teaching, and it is worth naming as a gift rather than a threat.
It is also not optional. An institution or a teacher who does not make this turn will not be punished for it; they will simply find themselves delivering something nobody needs delivered. Where a school is bound to a national examination that still rewards recall, it is not free to move quickly — but the direction holds, and arriving later is not the same as not going.
The world has changed. Teaching is how we answer.
Artificial intelligence is changing the world our students will work in, and it will not wait for us to be ready.
It is the school's job to teach students to use it so that it builds them up and does not tear them down — so that it strengthens their thinking rather than replacing it, and serves who they are becoming rather than hollowing them out.
What is scarce now is judgment — knowing what to do when no one has spelled out the answer, sensing that the problem in front of you is not the real problem, asking the right question in the first place. Judgment is the thing artificial intelligence supplies least, and so it is the thing a school must now grow on purpose. Building students up means growing their judgment. Tearing them down means letting the machine think for them until they have none of their own.
And judgment has to be aimed. Sharp sight is not the same as good sight — a swindler sees around corners too. To sharpen students' sight without aiming it is not a neutral act; it hands them a faster engine and no steering. So this policy asks for more than capable students. It asks that their capability be pointed at what is genuinely good, and that is not something a school can manufacture in them. "The inspiration of the Almighty gives understanding" (Job 32:8).
That is the whole aim, and everything that follows is measured against it.
What follows is a guide, not a finished policy. Each school has to think its own policy through and write it, because the people who will teach it and stand behind it are the ones who must decide it.
The guide holds two kinds of material, and they are meant to be used differently. Some of it is wording ready to be used — text a school can take as it stands, or adapt, and adopt as its own policy. The rest is matters to think through — questions where the right answer depends on the school, its students and its circumstances, and no one outside it can settle them.
Each entry below is self-contained, so a school can take one at a sitting and settle it before moving on — the contents at the top of this page are the way in. The reasoning underneath the whole thing, which the entries assume rather than repeat, is set out separately: why build one at all.
What counts as a student's own work?
Every other rule in a school's policy refers back to this one. Until it is answered, nothing else can be applied.
The usual answer is to ask whether the student used artificial intelligence. That question cannot be answered — not reliably, not by a teacher, and not by detection software, which is wrong about honest students often enough to be worse than useless, and is far more wrong about some students than others. The evidence for that is set out in the fourth entry.
There is a better question, and it can be answered in a two-minute conversation. Not "did you use the machine?" but "where are you standing?" There is more than one way to work with artificial intelligence, and the ways are not equal. They form a ladder.
The rung is not a property of the task. It is a property of the person doing it. Two students send the same request and get back the same work. One of them could have produced it unaided, knows exactly how, and chose to spend five minutes rather than five hours on something that would have taught them nothing they did not already have — that student is Guiding, and a school should say so approvingly. The other asked because they could not have produced it, and the gap is now hidden inside a finished piece of work. Same request, same output, opposite ends of the ladder. This is why "is using it allowed?" is the wrong question, and why a rule written about the tool can never answer it.
That is the whole test, and it turns the problem around. A school does not have to police what happened at home. It has to ask, in the room, whether the student can stand on rungs three, four and five. A student who can explain the work, check it, and say why it was done that way has done the work — whatever helped them produce it. A student who cannot has not, whatever they typed themselves.
Work submitted as a student's own must be work that student can comprehend, verify and guide. Artificial intelligence may be used in producing it.
What is not acceptable is submitted work that a student cannot explain in their own words with the machine switched off, cannot check for error, and did not direct.
A teacher may at any time ask a student to explain submitted work without the machine present. That explanation, and not the document, is the work being assessed.
- Which rung you require changes with the subject and the age of the student. A senior essay may reasonably demand Guiding. A young child copying letters is doing something else entirely, and the ladder does not yet apply to them. Decide where in your school it starts.
- Name the tasks where Doing is allowed. Sometimes a thing simply needs to be produced, and pretending otherwise makes the policy dishonest. A school that says so plainly — and says which tasks those are — is believed on everything else it says.
- Decide how the conversation happens, because the format is the whole difficulty. Asking a student to account for their work standing in front of the class puts the ones who leaned hardest on the machine in front of the people whose opinion they care about most, and they will not answer honestly. Done the other way — the teacher moving around the room while everyone works, a quiet word beside the desk — it costs no separate lesson and humiliates nobody. On where the time comes from, see a policy about the machine turns out to be about teaching, above.
- Decide before you need it what a teacher does when a student cannot explain their work. A school that has not settled this in advance will reach for punishment, because that is the only procedure it has. Settle an order of inquiry instead, and ask the questions in this order:
Did I teach this clearly?
Some of what a teacher says never arrives — the best any two people manage is about eighty per cent of what passed between them, and it falls to fifty or sixty across a culture, and to thirty when the lesson is not in the listener's first language. This is also the only one of the three a teacher can act on directly, which is why it goes first.Is something happening to this student?
A death in the family, work at home, a learning need nobody has yet identified. None of it announces itself, and none of it will be found by a question about artificial intelligence.Has this student not done the work?
A real answer, and sometimes the right one. It goes last because it is the one a teacher cannot un-ask.
The Freedom Ladder is from Original Intelligence by Dr Anthony J. Petrillo, in the chapter Working with AI. The chapters What AI Is — and Can You Trust It? and Does the Machine Have a Worldview? set out why the Verifying rung matters so much: the machine hands back the words people usually say, which is not the same as what is true, and it leans toward whoever is talking to it. The chapter One in Five is the source for the order of inquiry above — it gives the research on how much of what we say actually arrives, and how much further it falls across a language or a culture.
When must a student say they used it?
This is the only rule in the whole guide that can be taught rather than enforced — which is exactly why it has to be got right.
A school cannot detect use. It can build a habit of declaring it. Those are different projects, and only the second one is available.
But a declaration habit is fragile, and there is one thing that destroys it. If declaring costs the student anything, students stop declaring. A school that asks for honesty and then marks down the work that gave it has not taught honesty; it has taught concealment, and it has taught it faster and more thoroughly than any lesson on the subject. This is decided the first time a teacher marks a declared piece more harshly than an undeclared one — whatever the policy document says.
So the rule has to be stated the other way round, and stated plainly enough that students believe it: using the machine is not a breach of this policy. Concealing it is.
The second thing to get right is what a declaration should contain. "I used AI" is worth nothing — before long every student will write it, and it distinguishes no one from anyone. What carries information is what it did, and what you did. Those two clauses place a student on the ladder without anyone having to mention rungs, and they give a teacher something to ask a real question about.
Where artificial intelligence has been used in work a student submits, the student states briefly what it did and what they did. One or two sentences is enough.
Declaring use is never itself penalised, and work is not marked more harshly for having been declared. Work is assessed on what the student demonstrates. Using the machine is not a breach of this policy. Concealing it is.
A teacher may ask for more detail on any piece of work, and may ask the student to explain the work without the machine present.
- Whatever you decide, the students will find out what is actually true. They will compare marks. If declared work quietly does worse, the policy is dead within a term and nothing written down will revive it. Decide whether your staff can genuinely hold to this before you publish it.
- Do not ask for a declaration on everything. A student required to declare on every scrap of daily work will write the same sentence without thinking, and the habit you wanted becomes a formality that teaches the opposite. Decide which work carries it — assessed pieces, most likely — and leave the rest alone.
- Decide what a declaration looks like for your youngest students. A child who writes with difficulty can still say out loud what the machine did for them. If your policy assumes a written note, it has quietly excluded the students it most needed to reach. Spoken declarations are taken beside the desk during working time, not performed in front of the class.
- Ask what the school does with declarations once it has them. If nobody ever reads one or asks about one, students learn that too. The declaration is the opening of a conversation, and a school that never has the conversation has added a form to no purpose.
The question underneath a declaration is put directly in Original Intelligence, in the chapter Working with AI: "What have I already let the machine decide for me — and could I still do it without it?" A student who can answer that in a sentence has done the thing this entry is really asking for. The declaration is only the place where it becomes visible to a teacher. The book can be downloaded at baba.computer.
What may faculty use it for in their own work?
A guide written to faculty that governs only students has a hole in it, and the students will find the hole before the staff do.
Teachers are using it. To plan lessons, to draft worked examples, to reword an explanation that did not land, to get through a pile of reports at eleven at night. A policy that does not say so is not describing the school it was written for.
And the reason to settle it is not fairness in the abstract. A teacher who requires of students what they do not practise themselves will be found out — a comment in a corridor, a document with the wrong tone, a student who recognises the phrasing. When that happens, everything the school has said about artificial intelligence becomes negotiable at once, including the parts that mattered.
The line worth drawing is not between using it and not using it. It is between preparation and judgment.
That test is the same rule the guide already asks of students, pointed the other way. If a teacher would not be comfortable disclosing it, the school has learned something it should not ignore. Reciprocity is not a courtesy here — it is the only thing that makes the disclosure rule in the previous entry survivable.
One more thing belongs to faculty rather than to students. A teacher is on the same ladder. A lesson prepared at the bottom rung — produced, accepted, not understood — holds together exactly as long as nobody asks a question that steps off the script. It has never been easier to stand in front of a class holding an answer you do not understand.
Teachers may use artificial intelligence in preparing to teach: planning, worked examples, practice material, rephrasing an explanation, translating or simplifying a text, and getting the wording right in a language that is not their first.
A mark, and any comment written on a student's work, is a judgment the teacher makes and stands behind. Where artificial intelligence has helped produce it, the teacher has checked it, the students have been told, and any error in it is resolved in the student's favour. The same holds for reports and references.
Before using it on anything concerning a particular student, apply one test: would you be willing to tell that student exactly what you did? If not, do not do it.
Students' work and students' personal details are not entered into any service the school has not approved.
- Machine-assisted marking is workable, and there is a shape that makes it safe. One teacher's practice, offered as a worked example rather than a recommendation: the machine produces a first pass over the work, saying where the student did well and where to focus; the teacher then reviews it as the expert, which is where a machine's read gets corrected. Three conditions carry it. The students are told plainly that this is how their work is marked. They are asked to check the marking, and if they find an error they get the credit. And if the error ran the other way and the machine was too generous, they keep that too. Error can then only ever move in the student's favour, so no student is harmed by a machine mistake — and the students spend that time reviewing their own work against a critique, which is the Verifying rung being practised on the one piece of work they care most about. Whether your school can run this depends on staff who will genuinely do the review; a first pass nobody checks is not this practice, it is abdication wearing its clothes.
- Settle reports and references now, in a calm week. They are the hardest case in the whole guide — the volume is crushing, the temptation is highest, and they are the documents that follow a student longest. A school that has not decided this before report week has decided it by default, at eleven at night, one exhausted teacher at a time.
- Say out loud that language help is allowed. Where the machine helps a teacher most is often where that teacher is weakest — writing in a second language, or teaching a topic at the edge of their training. That is legitimate use and it should be named as legitimate, or it will simply be hidden, and hidden use cannot be guided.
- Student work leaving the building is not a matter of school preference. It is law, and the law differs by country. Pasting a student's essay or name into a service the school has no agreement with sends an education record to a third party. In the United States that is governed by FERPA (the Family Educational Rights and Privacy Act — the federal law protecting the privacy of student education records), and the violation is committed by the institution, not by the teacher who pasted. FERPA does allow an outside service to be used, but only where the school has designated it a school official with a legitimate educational interest, keeps direct control over how the data is used, and confines it to that purpose. That is exactly what "a service the school has approved" has to mean — a real agreement, not an assumption. Every country's rule is different. Find out which law binds your school, name it in your policy, and have someone qualified read what you wrote. Do not copy this paragraph and hope.
- Ask whether the school will apply the disclosure rule to itself. If students must say what the machine did, the honest position is that teachers say so too — at least to each other. Deciding not to is defensible; deciding it quietly is not, because students will work out which rule the school actually holds.
The chapter Knowledge — the Good Gift That Puffs in Original Intelligence is written about students but reads harder on teachers: it is possible to know a subject completely and be no use at all to the person sitting across the table. The machine can supply the knowledge. It cannot supply the person who knows the student, and that is the part of a teacher's work this entry is protecting.
What do you set, now that the machine can do the old assignment?
If a task can be completed in full by artificial intelligence, it has stopped telling you anything about the student who handed it in. That is a fault in the task. Punishing the student who used the machine does not repair it.
The instinct is to make the task harder to complete with a machine — detection software, locked-down rooms, everything handwritten. Every one of those is an arms race, and the school is not going to win it.
Worse, they change what a school is optimising for. A teacher who sets work to be machine-resistant is no longer setting it to be instructive, and those two come apart quickly. The question to ask about a task is not whether a machine could do it. It is what the task was supposed to reveal about the student, and whether it still reveals it.
The ladder gives the test. A task that can be completed standing on the bottom rung measures nothing, because a student who understood the material and a student who did not will hand in the same thing. Set work that cannot be finished without comprehending, verifying or guiding, and the difference becomes visible again — without anyone having to police anything.
None of this means every task must defeat the machine. Some work exists to be produced and got out of the way, and the first entry says so plainly. The requirement is not that tasks be machine-proof. It is that a teacher knows which kind each task is, and stops awarding marks for the kind that no longer measures anything.
Where a task can be completed in full by artificial intelligence, the task is revised rather than policed.
Assessed work should require the student to do at least one of the following: account for the route they took, work with material that exists only here, find and correct an error in what a machine produced, or explain the work aloud.
Teachers are not asked to make work resistant to artificial intelligence. They are asked to know what each task measures, and not to award marks for tasks that no longer measure anything.
- Do not start an arms race, and understand what detection software would cost you before you buy any. In 2023 Stanford researchers ran ninety-one essays written by human students whose first language was not English through seven leading detectors. Sixty-one per cent were flagged as machine-written. Eighty-nine of the ninety-one were flagged by at least one detector, and eighteen were flagged by all seven at once. The same detectors almost never made that mistake on the writing of native English speakers. The reason is mechanical rather than malicious: a student writing in a second language uses the careful, formal, predictable structures they were taught, and predictability is the very thing these tools read as machine-written. For a school teaching in English to students who speak something else at home, detection software is not unreliable. It is systematically wrong about exactly your students, and every accusation it produces will land on the child who worked hardest at their English. Vanderbilt University turned Turnitin's detector off in 2023 for this reason. Lock-downs are cheaper and still cost you something — they teach a class that the school's first assumption about them is dishonesty.
- Where a national examination still rewards recall, you cannot stop preparing students for it. Their futures turn on that paper and no policy overrides it. But the examination is a week and the year is not — the classwork can change long before the syllabus does, and students who understand will do better on a recall paper anyway.
- Work that shows the route takes longer to mark than work that shows only the answer. That is real, and it is the same reallocation the teaching section describes: the hour comes from delivery, and a teacher who is still lecturing knowledge has not freed it yet. Do not adopt this while leaving the lecture in place.
- Say which tasks are still meant to be produced quickly. If a school never names them, teachers will assume every task must now be elaborate, and the workload objection will kill the whole guide within a term.
The chapter Working with AI in Original Intelligence opens with the student this entry exists to catch: bright, fast, everything handed in for a year looked excellent — and then, given a problem simpler than anything she had done all year, she could not do it. Nothing in her submitted work would have revealed that. The same chapter notes that the gap was already open before the machine arrived: a 2003 study of adult literacy found the proportion of college graduates reading at a proficient level had fallen to about a third. They could read and repeat. What they could not do was follow an argument and say what it meant.
What about students who cannot reach it?
A policy written as though every student has a device and a connection will quietly award marks for their families' means. Nobody intends this, and it happens anyway.
The rule that prevents it is short. Nothing a school assesses should require access the school does not itself provide. Everything else in this entry follows from that one line.
It cuts both ways, and schools usually see only one of them. Work that expects artificial intelligence punishes the students who cannot reach it. Work that forbids it advantages the students who can — because they can use it and nobody will know. The same policy can be unfair in both directions at once, in the same classroom, on the same afternoon.
But the harder point is the one that gets missed entirely, and a school should decide it with its eyes open rather than by default.
This is not an argument for permitting everything — the whole guide is about teaching students to climb rather than to lean. It is an argument for knowing what a ban costs and to whom, before writing one.
And one practical thing is worth saying plainly to any school reading this without a reliable connection. Teaching students to work with artificial intelligence does not require a device for each student. The most valuable lesson in this guide — here is what the machine produced, find what is wrong with it — needs one printout, or one screen at the front, or a teacher reading the machine's answer aloud from a page. Judgment is what is being taught, and judgment does not need bandwidth.
No assessed work requires a student to have access to artificial intelligence outside school. Where a task depends on it, the school provides the means during school time.
A student without a device or a connection is not disadvantaged in any mark, and is never asked to explain why they lack one.
Teaching students to work with artificial intelligence does not require a device for each student, and the absence of one is not a reason for a school to leave this untaught.
- Access is not a yes or a no, and a policy written as though it were will mis-sort your students. A student may have a phone but pay by the megabyte. A student may reach it only when a parent is home in the evening. Several children may share one device and one account. Ask what access actually looks like in your students' homes before you write a rule that assumes it.
- Decide what you think about shared accounts. A student working on a parent's or an older sibling's account leaves their schoolwork in someone else's history, and reads whatever is already in it. This is a safeguarding question more than a technical one, and it is more common than any school's own survey suggests.
- Do not let "we have no connectivity" become "this does not concern us." The students will not stay inside the school. They leave for cities, for work, for further study, and for phones they buy themselves, and the machine will reach every one of them. What a school controls is not whether they meet it, only whether they meet it having been taught — or meet it alone, as adults, at the point where trusting it wrongly costs a job or a sum of money the family cannot spare. And what a school can teach without any connection is the half that lasts. Which tools exist and which buttons to press will have changed before this year's class has finished leaving. Knowing that a confident answer can still be false, and having practised finding out, will not. A school with no connection can still teach the durable half; a school with a fast connection can spend a whole year teaching the disposable half and call it preparation.
- Beware the reverse assumption too. A school in a well-connected city can slide into requiring the machine for everything, and then a student whose family has lost its connection this month is quietly failing for a reason nobody has asked about.
The chapter Working with AI in Original Intelligence makes the point that this is not confined to schoolwork: "It is not just students, and not just schoolwork. It is any of us — with a recipe, a letter from a lawyer, a doctor's word, a decision about money — letting the machine hand us the answer while we quietly grow emptier." The students for whom the machine is the only adviser available are the ones who will lean on it hardest for the rest of their lives, and so they are the ones who most need to be taught to climb.
What may students put into it?
Students treat it as a private notebook. It is not one. Nobody has told them, and the school is the only party in a position to.
Everything typed into artificial intelligence leaves the building. It is held by a company, may be kept, may be read by a person, and may be used to train the next version. A student typing at eleven at night has no sense of any of that — the screen looks like a diary and answers like a friend.
Three different things are at stake, and schools usually think of only the first.
The third one deserves more than a line, because it is the one that can end badly. A distressed student talking to a person meets someone who can be alarmed, who can stay, who can fetch help. A distressed student talking to the machine meets something that will agree with them. It leans toward whoever is speaking to it; hand it your account of your own situation and it will hand the account back, tidier and more confident. For a student who is frightened, or ashamed, or being harmed, agreement is precisely the wrong response — and it arrives instantly, at any hour, with nobody ever finding out.
So a school's rule here is not really about data. It is about making sure students know what the machine cannot do, and know the name of a person who can.
Before students use artificial intelligence in school, they are taught that everything typed into it leaves the school, is held by a company, and may be kept and read.
Students do not enter their address, their photograph, their health or their family circumstances. They do not enter another person's name or circumstances at all: a student may give away their own privacy, but not anybody else's.
A student who is worried, frightened or in trouble is directed to a person. The school makes plain, in words students remember, that the machine cannot notice distress, cannot act on it, and cannot tell anyone — and names the members of staff who can.
- Teach this before permitting use, not after an incident. It is one lesson, and it is the only part of this entire guide that has to happen in a particular order. A school that permits first and explains later has already sent whatever was going to be sent.
- Name the people, do not name a category. "Talk to a trusted adult" is advice a frightened student cannot act on, because working out who qualifies is the very thing they cannot do at that moment. Two or three names, said often enough to be remembered without thinking, is a different instruction entirely.
- Check the age terms of whatever your students are using. Most of these services set a minimum age, and some require parental consent below another. A school that directs a class to a service its students are too young to hold an account on has taken on a problem it did not know it had, and the families will hear about it from someone other than the school.
- If the school provides the access, decide what the school can see — and say so first. An account the school controls may leave every conversation visible to staff. That may well be the right arrangement. Discovering it afterwards is what students will not forgive, and it will end their honesty about everything else in this policy.
The chapter Does the Machine Have a Worldview? in Original Intelligence names the mechanism behind the third danger above. The machine leans toward whoever is talking to it: "Tell it where you stand, hand it your own way of seeing, and it will happily reflect it straight back — agreeable, confident, tidy. That can be genuinely useful. It can also be a mirror that flatters." And then the sentence a school should keep in mind when a student is alone with it at night: "A thing that feels neutral, never gets tired, and gently agrees with you is the easiest mind programmer you will ever let into the room."
How do students check what it tells them?
"Check it" is not teachable advice. It names the duty and withholds the method, and a student who does not know how will simply decide they have checked.
The machine is wrong sometimes, and it is wrong in a particular way that matters: it is wrong in the same calm, complete, confident voice it uses when it is right. There is no tone change, no hedge, no tell.
That is not a defect someone will fix next year. It follows from what the thing is. It was built to produce the words that usually follow — and what people usually say is not the same as what is true. Where the words it learned from were wrong, it hands the error back politely. Where there were no words at all, it fills the gap with something plausible and states it with a straight face.
A teacher who cannot explain that will lose the argument to the first confident student in the room, which is why this entry belongs to faculty before it reaches students. The explanation does not require any technical background: the machine has read more than any of us could read in a thousand lifetimes, and what it learned was not what any of it meant — only which words tend to follow which. It is an extraordinary guesser. It is not a knower.
A short example makes the middle one teachable, and it happened to the author of the book this guide draws on. He fed one of his own pieces of writing into a tool that promised to use only the material given to it. When the machine read his work back, it had inserted four words that were not his: "in the opinion of the author." Nothing had been added and nothing removed — but something he had stated as true had been quietly returned to him as a personal opinion. Nobody edited it. The crowd it learned from simply talks that way, and the softening was delivered as neutrality.
Which gives the method. Verification is not a separate subject; it is the top of the ladder and it rests on the rung below it. You can only check what you comprehend. A student who cannot say the thing in their own words has nothing to compare the answer against, and their "checking" will amount to reading it again and finding it convincing — which is what the machine was optimised to be.
Students are taught that artificial intelligence produces the words that usually follow, which is not the same as what is true, and that it states a falsehood in the same voice it uses for a fact.
Where a student submits a claim that came from a machine, they are expected to have checked it and to be able to say how they checked it. Naming a source is not checking it: the source must exist and must say what it is said to say.
Verification is taught as a skill and practised in lessons. It is not issued as an instruction and left to the student.
- Teach a method, not a duty. Something a student can actually run: say it in your own words first; ask whether it contradicts itself; ask whether it fits what you already know to be true; go and see whether the source exists and says that; put the same question a second time in different words and see whether the answer holds. Five things a fourteen-year-old can do. "Be critical" is not one of them.
- Most verification failures are comprehension failures wearing a disguise. When a student cannot check something, the useful question is not whether they were careless but whether they ever understood it. That points the teacher at the real gap, and it is the same diagnosis the first entry asks for.
- Slant is harder to teach than error, and it matters more. An answer can be entirely accurate and still arrive carrying assumptions about what is normal, what is settled, and what is merely somebody's opinion. Decide whether your school will teach students to ask where is this answer standing? — and be aware that a school choosing not to has not thereby stayed neutral.
- Watch for agreement in your best students, not your weakest. The student most likely to be flattered by the machine is the one confident enough to tell it what they think before asking. They will come back with their own view, better argued, and mistake that for having tested it.
Two chapters of Original Intelligence sit directly under this entry. What AI Is — and Can You Trust It? explains what the machine actually does, in terms any member of staff can repeat to a class, and lands on the rule this guide follows: "we do not hand it our thinking. We use it, and we check it." Does the Machine Have a Worldview? is the source of the example above and of the harder half of this entry — the machine holds no worldview of its own, which is exactly why it will hand a student somebody else's without either of them noticing.
What happens when the policy is broken?
This entry decides whether the seven above it were meant. A guide that teaches for seven entries and punishes in the eighth was a fence all along, and everyone will know it.
A school that has not settled this will reach for the procedure it already owns — the one written for plagiarism. That is the wrong instrument, and reaching for it is the single most common way these policies fail.
Plagiarism is passing off a person's work as your own. There is an author who was wronged and a deliberate deception about who wrote something. Most of what a school will actually meet here is not that. It is a student who produced something they never understood — and the remedy for not understanding has never been a penalty.
So separate the two things that arrive looking identical, because a school that cannot tell them apart will treat them identically and be wrong about half of them.
Two things follow, and both protect teachers as much as students.
Nobody is accused on the strength of detection software or a teacher's sense of style. The figures in the fourth entry are the reason: these tools are wrong about second-language writers most of the time, and a teacher's ear for a pupil's voice is not evidence — it is a hunch about a fifteen-year-old who has been reading. The evidence this guide relies on is the same from the first entry to the last: can the student account for the work? That question is answerable, it is fair, and a student who can answer it has ended the matter.
And a school should be plain with itself about concealment, the second of the two above. Treating deliberate concealment as though it were only a learning need is its own kind of dishonesty, and staff will quietly stop believing a policy that asks them to pretend. The guide has argued throughout for teaching rather than fencing. It has not argued that nothing is ever wrong.
Where a student cannot account for work they have submitted, the first response is teaching. The work is done again, with the teacher, until the student can explain it. This is not a disciplinary matter and is not recorded as one.
Where a student has denied using artificial intelligence when asked directly, or has taken steps to conceal it, that is a question of honesty and is handled as this school handles honesty anywhere else.
No student is accused on the basis of detection software, or of a teacher's impression of how the writing sounds. The evidence is whether the student can account for the work.
- Decide what happens the second and third time, before the first time happens. A policy with a single response is either too heavy for a first occasion or too light for a pattern, and whichever it is, a teacher will discover it at the worst moment. A pattern of a student who never climbs is a different problem from a single lapse, and it usually means something the school has not yet found out.
- This protects teachers more than it restrains them. One accusation that turns out to be wrong costs a teacher the trust of a class for years, and the class will remember it long after the student has left. A rule that requires the conversation before the accusation means no teacher ever has to rely on a hunch in front of a parent.
- Decide how a student disagrees with you. A student who did understand the work, and froze when asked, needs a way to say so and be heard — a second conversation, another member of staff, a day's grace. Schools that skip this are the ones that produce the case everybody still talks about five years later.
- Read the first entry again alongside this one. The order of inquiry there — did I teach it clearly, is something happening to this student, has this student not done the work — is the procedure this entry assumes has already been followed. Without it, the whole thing collapses back into catching people.
The chapter Working with AI in Original Intelligence states the aim that every response in this entry is meant to serve: "The trap is not using AI. The trap is living on the bottom rung." A school's answer to a breach is judged by one thing — whether it got that student off the bottom rung, or merely recorded that they were standing on it. And the question the same book puts to students is the one a school should keep in view while writing any of this down: am I becoming someone worth trusting with what I know? That is what the policy is for. It was never about the machine.
Notes on sources
Detection software and second-language writers. The figures in the fourth entry come from a 2023 study by researchers at Stanford University, which passed ninety-one TOEFL essays written by human students through seven widely used AI detectors: 61 per cent were classified as machine-generated, 89 of the 91 were flagged by at least one detector, and 18 were flagged unanimously. Comparable essays by native English speakers were rarely misclassified. Vanderbilt University disabled Turnitin's AI-detection feature in August 2023, citing both the false-positive rate and this bias. The study has been disputed, most vocally by companies that sell detection software; the mechanism it describes — that predictable, carefully learned prose is what these tools read as machine-written — has not been shown to be wrong.
The Freedom Ladder, and the chapters cited throughout, are from Original Intelligence by Dr Anthony J. Petrillo, used with the author's permission. The book can be downloaded at baba.computer.
Student data. The account of FERPA in the third entry describes United States federal law, including the school-official exception at 34 CFR § 99.31(a)(1). It is offered as an illustration of the kind of law that applies, not as legal advice, and it does not apply outside the United States. Every school should establish which law binds it and have someone qualified read what it writes.
The guide stays open. If your school reaches a decision this does not cover, tell us and it will be worked through and added to the next edition. Get in touch.