AI in Learning:
Policy and Guidelines

This policy sets out how AI may be used in coursework and assessment, how that use is disclosed, and how questions about authorship are resolved.

It starts from the position that AI is now part of how people learn and work, and that a school's job is to teach students to use it well rather than to contain it. The test applied throughout is not whether AI was used, but whether the student still did the thinking the course was designed to teach.

Three commitments run through every section.

Transparency AI use is declared, not hidden. Undisclosed use is the problem. Disclosed use within stated limits is not.

Human authorship A person is accountable for everything submitted under their name, including AI-generated errors, fabricated sources, and bias.

Skill protection No use of AI may come at the cost of the skills a student is here to develop. This document is a template. Bracketed fields mark where a school inserts its own detail. Because technology moves faster than any policy, it is written as a starting point rather than a settled answer, and is revised openly with students at the table.

Three things distinguish it.

1. It is operational. Every stage of an assessment carries a stated level, every level has a worked example, and disclosure has a form rather than a principle. 2. It assumes no detection software, because detection does not work (Section 9) and a policy built on it collapses the first time it is challenged. 3. It is portable. It names no curriculum, no jurisdiction, and no products, so a school in any system can adopt it by filling in the bracketed fields.

This policy applies to all students, teaching staff, and support staff at [School Name], and covers all generative AI tools (text, image, audio, video, and code) used for coursework, assessment, school-related communication, and school administration.

It sits alongside, and does not replace, [school's academic integrity policy], [school's acceptable use policy], and [school's data protection policy]. Where this policy and a teacher's written assignment instructions differ, the assignment instructions apply.

Equity of access. No assessment may require a tool that not every student can access. Where a tool is required, the school provides it or provides an equivalent alternative. This applies regardless of level.

1. Transparency. Undisclosed AI use is the problem. Disclosed use, within the limits set for a task, is not. Where a student is unsure whether something needs disclosing, disclosing it is never the wrong answer. 2. Human authorship. A student is accountable for everything they submit, including AI-generated errors, fabricated sources, and bias. "The tool wrote it" is not a defence and never has been. 3. Skill protection. AI may support how a student learns. It may not replace the learning itself. The same holds on the staff side. Decisions about grades, reports, references, admissions, and discipline are made by people, because the judgment is the job.

AI permission has two parts, the stages an assignment is divided into and the level set for each stage. Section 3.1 sets out how stages are defined. Section 3.2 defines the four levels.

AI permission is set per stage of an assignment, not for the assignment as a whole. A stage is a distinct phase of work, such as research, idea development, production, or evaluation.

Every assignment that permits AI use at any stage defines its stages in advance and states the permitted level for each. Stage definitions follow the subject’s existing assessment structure where one exists. [Design uses criteria A to D. Sciences use the internal assessment phases. Other subjects define stages at department level.]

The permitted level for a stage is determined by the skill being assessed at that stage. Where the assessed skill is one that AI can perform, the stage is Level 0. Where the skill is one AI cannot perform, the stage may be opened up, and the restriction spent elsewhere.

A single assignment will normally use more than one level. An assignment set entirely at one level is usually a sign that the stages have not been examined.

Example. One assignment, four stages, four levels. A Grade 10 design project in which students develop a stackable stool.

The same logic applies to any assignment with a defined process. A written assignment might permit AI while a student builds understanding of a text, prohibit it while the thesis is formed, and permit it again during revision.

Each level is set out the same way. The rule for students is the sentence a student reads for that stage, worded identically wherever this policy is used, including on rule cards and in any app that carries it. The fields below it

say what AI may touch, what the student produces unaided, what goes in the process journal, and what is disclosed. The level is stated in writing by the teacher who sets the assignment, against each stage.

Related framework. The AI Assessment Scale (Perkins, Furze, Roe and MacVaugh, 2024, aiassessmentscale.com) is a similar level- based framework.

Rule for students: Do this stage without any AI tool.

AI may touch: nothing at this stage, including planning and proofreading.

Student produces unaided: all of it.

Journal entry: required. It may be a single line confirming no AI was used, with anything the student wants to say about the hardest part.

Disclosure: the stage is listed in the statement as completed without AI.

Applies to: a stage, not an assignment. A single assignment may use several levels.

Example: the evaluation stage of a design project, or a timed in-class essay set as a single stage.

Rule for students: Use AI to explain, translate, or check comprehension. Nothing it produces enters your work.

AI may touch: the student's understanding around the work, such as explaining a concept, translating a source so the student can read it, generating practice questions, or checking that the student has understood the material.

Student produces unaided: every word, line, and calculation submitted. Nothing AI produces enters the work.

Journal entry: what the student asked AI to explain, and what they understand now that they did not before.

Disclosure: the tool and what it was used for, with confirmation that nothing it produced entered the work.

Applies to: a stage, not an assignment. A single assignment may use several levels.

Example: a maths problem set where a student asks AI to explain a method and generate five similar practice problems, then solves the assigned questions alone. Asking AI to check a submitted answer is Level 2, not Level 1. The line is whether AI touched the work being handed in.

Rule for students: Generate, draft, or critique with AI. Log it. The final decisions must be yours.

AI may touch: the work itself, by generating options, drafting, or critiquing. This covers brainstorming, a first draft of a section, structural feedback, translation, editing, and code debugging.

Student produces unaided: the final decisions, meaning which AI output is kept, rewritten, or rejected, and why. The student is accountable for every decision in the submitted work.

Journal entry: what AI gave, what the student kept, rewrote, or rejected, and why.

Disclosure: required (Section 5), naming the stage, the tool, the purpose, and what the student kept, changed, or rejected.

Applies to: a stage, not an assignment. A single assignment may use several levels.

Example: the idea development stage of a design project, where a student generates eight possible forms with AI, tests them against the brief, keeps two, and explains why the other six were rejected.

Rule for students: Use AI freely here. Log what you used and reflect on what it changed.

AI may touch: anything at this stage.

Student produces unaided: the reflection on what AI changed, and responsibility for the accuracy of everything used.

Journal entry: what AI did at this stage, and what would have been different without it.

Disclosure: required (Section 5), listing the tools used and what they changed.

Applies to: a stage, not an assignment. A single assignment may use several levels.

Example: the research stage of a design project, where research synthesis is not the skill being assessed, so students use AI freely to gather and summarise background material and then reflect on what it changed.

Default rule. Where no level is stated for a stage, Level 1 applies. Level 1 rather than Level 0, deliberately. A Level 0 default would make the most ordinary uses of AI, such as looking up a definition, generating practice questions, or asking for an explanation of something taught badly the first time, a violation by silence. Students would breach the policy daily without knowing it, and would learn to treat the whole document as unrealistic. Level 1 puts the default line where the actual risk begins, at the point AI produces submitted work. Where an assignment defines no stages at all, it is treated as a single stage at Level 1. Students who are unsure must ask before starting, and teachers must answer clearly. A teacher who has not stated a level cannot treat Level 1 use as misuse.

Section 9 rules out detection. This section is what replaces it.

A policy that cannot verify authorship by inspecting a finished text has to verify it somewhere else. There are two places, work done live in front of people, and work whose process is recorded as it happens. Neither is surveillance. Both are things good teaching already does. This policy just makes them deliberate rather than incidental.

Some part of every unit is assessed in the room, such as a short presentation, a demonstration, a seminar contribution, a whiteboard walkthrough, or a two-minute oral follow-up on a submitted piece.

The design rule is that no student's grade for a unit should rest entirely on unsupervised text. Where a skill genuinely matters, it is assessed at least once in a setting where AI cannot be present.

This does more than confirm authorship. A student who can explain their argument, take a question on it, and extend it under mild pressure has demonstrated the learning directly, which is what the assessment was for. It also makes Section 9 workable. When a teacher has a question about a submission, the conversation is a normal continuation of how the course already runs, not an accusatory departure from it.

Live components are short. A two-minute oral follow-up on a major essay is sufficient. A second full assessment is not the intent.

For substantial work done outside class, students keep a process journal with one entry for every stage, including stages at Level 0. The entry is made when the stage is done, not reconstructed at the end. The guide is at Annex C.

An entry is spoken or written. A spoken entry is up to sixty seconds, said aloud and converted to text as the student speaks. Only the text is kept. No audio recording is made or stored. Before saving, the student may correct words the conversion got wrong, and a corrected entry is marked as edited. A written entry is available to any student as an equal alternative, at any time, and no reason need be given for choosing it.

An entry says what the student did, what AI contributed if anything, and what the student decided. A stage completed without AI is recorded as such. The question the student answers depends on the level of the stage (Annex C), and the teacher may add one question of their own for each stage.

Where physical work is the point of a stage, such as making, art, or lab technique, the teacher may ask for a photo of the work alongside the entry. Photos show the work, not the student.

Students may speak or write in any language in which they are fluent, unless the assignment assesses production in a specified language. Journal entries [count toward the process criterion of the relevant assessment / are required before work is accepted for marking].

Speech is the default because a written reflection can be produced by the same tool the student is reporting on. A student explaining their own decisions aloud cannot be, at least not without more effort than doing the work honestly. Converting speech straight to text keeps that benefit without keeping a recording of a student’s voice.

Journals are formative. They are read to understand how a student thinks, to intervene earlier when someone is stuck, and to give the student a record of their own reasoning to draw on.

A journal is not an evidence file. It may not be trawled for inconsistencies to build a case, and a thin or missing journal is not by itself grounds for an authorship finding. Some students think in their heads, some had a difficult week, and neither is misconduct. Where a journal contributes to a Section 9 conversation, it does so as one input the student is shown and invited to speak to.

Privacy. No audio or video recording of a student is made or kept for the journal. Entries are text, accessible only to the student and the teacher of that class, and retained with the coursework. Where an AI tool processes entries, for example to summarise them for the teacher or to compile the disclosure statement, the provider

handles the text without keeping it or using it for training, the same rule that applies to any speech-to-text tool. Entries are not used to assess spoken fluency, confidence, or language proficiency unless the assignment explicitly assesses those skills, and are not used for identification or for comparison between students. No student is required to speak in order to complete an assignment.

Equity note. Both arenas must be workable for students without quiet study space, stable internet, or the confidence to speak in front of a class. Live components should offer a low-stakes format alongside a public one, and a one-to-one walkthrough counts. Journal entries may be spoken, typed, handwritten, or bullet- pointed.

Every submission of a staged assignment carries a short AI use statement. It covers each stage and states

• the level each stage was set at;

• which tool or tools were used, if any;

• what they were used for;

• what the student kept, changed, rejected, or verified. Stages at Level 0 are listed as completed without AI. Where the student kept a process journal, the statement is compiled from their entries, then reviewed and signed by the student. A signed statement is not revised silently. Any later change is recorded as a new version, and both remain available to the teacher.

One statement covers the whole assignment. Disclosure is not a penalty and does not by itself lower a grade. Failure to disclose is treated as an integrity matter. The template is at Annex B.

This policy names no products. Tool availability differs sharply by country, by school licensing, and by network access, and a policy that names tools ages faster than one that names categories. What is approved is therefore recorded locally, in Annex A, and reviewed each term.

Tools not listed are not prohibited by default, but a student using an unlisted tool must name it in their disclosure. Staff add tools to this annex. Students may propose additions at the annual review.

Regardless of level, no member of the community may

• submit AI-generated work as their own, or use AI at a stage set to Level 0;

• use AI to complete work on another student's behalf;

• enter another person's private information, medical or wellbeing information, or unpublished work into an AI tool without their consent;

• generate images, audio, or video depicting a real member of the community without their consent, or any content that harasses, demeans, or impersonates another person;

• present AI output as verified fact without checking it, particularly citations, quotations, statistics, and translations.

Students state the level they were working under at each stage, make a journal entry for each stage where journals are required, disclose as required, keep drafts or version history, and verify anything AI produces before relying on it.

Teachers define the stages of every assessment and state a level against each, design tasks that make the intended levels realistic, include at least one live component per unit, teach students how to use AI well rather than only what is forbidden, and review any AI-assisted feedback, comment, or report before it reaches a student or family. Teachers do not enter identifiable student work or personal data into tools not approved by the school.

The school maintains Annex A, ensures listed tools meet the school's data protection obligations and any minimum-age requirements, ensures any speech-to-text tool used for journals keeps no audio and any tool that summarises or compiles journal entries keeps no entry text, introduces this policy to staff and students at the start of each year, and funds access where a tool is required. Introducing the policy is a briefing, not a training programme. See Section 10.

Detection software is not used. Not "used with caution". Not used. No detector output may form the basis of a finding, sole or contributing, and no student may be asked to account for a detector score.

This is not a soft position. It follows from what the tools actually do.

• Text written before generative AI existed is routinely flagged as AI-generated. Historical documents, published literary work, and student essays written years before these tools were available all return high AI-probability scores. A system that flags a document written two hundred years ago is not measuring authorship.

• The same submission produces materially different scores across different detectors, and sometimes across the same detector on different days. A measure that does not agree with itself cannot ground a finding about a student.

• Detectors over-flag students writing in a second language and students who write in plain, direct sentences, who are disproportionately the students least equipped to contest an accusation. A school that acts on these outputs will be wrong about specific students, will be unable to explain why when challenged, and will be wrong most often about the students it can least afford to be wrong about.

Where a teacher has a question about the authorship of a submission, the first step is a conversation with the student, informed by their journal where one was kept. It is raised as an educational conversation, not an accusation. The student is told what prompted the question and given the chance to respond. Evidence is drawn from the conversation itself, drafts and version history, process journal entries where kept, the student's prior work, and the student's ability to explain and extend their own submission.

A first substantiated case normally ends in resubmission at the correct level, plus a documented conversation with the student and family. The resubmission is the remedy, not a penalty attached to one. The aim is that the student does the learning the task was set for. Repeated or deliberate misuse follows [school's academic integrity procedure].

Where the conversation does not resolve the question, no finding is made. Uncertainty resolves in the student's favour.

This policy is designed to be adopted in a staff meeting, not a training programme. If it needs a workshop to be understood, it is written wrong.

What each group actually has to do.

• Teachers. Write a level against each stage of an assessment. In most subjects the stages already exist in the assessment structure, so this is labelling phases that are already there, not inventing new ones. That is the whole obligation. Everything else in this document exists to make that one act meaningful.

• Students. Read three things, the four levels (Section 3), the AI use statement template (Annex B), and the process journal guide (Annex C). Fifteen minutes.

• Leadership. Fill in the bracketed fields and Annex A, name the stage structure each department will use, and add a stage-and-level table to the school's assessment template so the levels are structural rather than remembered. Four steps to adopt.

1. Adapt the bracketed fields and complete Annex A. One sitting. 2. Agree the stage structure per department, and add a stage-and-level table to assessment cover sheets and the LMS assignment template. 3. Run one department for one term with both arenas from day one, starting in a subject that already assesses process. A journal that is optional and ungraded will not be completed. Collect the questions teachers actually get asked. 4. Revise from those questions, then extend school-wide. Expect the first term to surface disagreements about where Level 1 ends and Level 2 begins in specific subjects, and about where one stage stops and the next starts. That is the policy working, not failing. Those disagreements are the ones worth having, and Section 11 is where they get resolved into the next version.

This policy is reviewed at least once per academic year by a working group that includes students, teaching staff, and a member of the leadership team.

Maintained by staff. Reviewed each term. Students may propose additions at the annual review.

Attach to every submission of a staged assignment. One statement covers the whole assignment. Where a process journal was kept, compile the statement from the entries. Two to four sentences is usually enough.

Assignment: [ ]

Stages and levels: [stage name, level 0 / 1 / 2 / 3, for each stage]

Tool(s) used, by stage: [name each tool, or "no AI" for the stage]

Used for: [e.g. explaining a method; generating design options; structural feedback on paragraphs 3–5]

What I kept, changed, rejected, or verified: [ ] Level 3 stages only, what AI changed: [ ]

Signed and dated: [ ]

Example. "At the research stage (Level 1) I used [tool] to explain two terms in my sources. Nothing it produced is in my work. At the drafting stage (Level 2) I used [tool] to give feedback on the structure of my argument after I had written a full draft. It suggested moving my counterargument earlier, which I did. It also proposed two sources I could not verify, so I removed them. The evaluation stage was done without AI."

One entry per stage, including stages at Level 0. Spoken, up to sixty seconds and converted to text, or written. Only the text is kept. Not marked for writing quality.

Stage: [ ]

Level: [0 / 1 / 2 / 3] AI tool used: [name, or none]

Levels 2 and 3, what I did with the output: [kept / rewritten / rejected]

The question depends on the level of the stage. It stays in front of the student while they speak.

Teacher's question for this stage: [one question, added by the teacher]

Photo of the work, where the stage asks for one: [attach]

After speaking, the student reads the text, corrects any misheard words, and saves. A corrected entry is marked as edited. The student may re-speak once.

Tool used: [name, or none]

What I asked: [ ]

What came back: [Levels 2 and 3, optional]

What I did with it, and why: [ ]

Level 1 only: [I confirm nothing AI produced entered my work]

Teacher's question for this stage: [ ]