P1-S6 School planning: whole-school AI risk register and incident response (eight columns, ten risks, three steps)

• StageP1-S6
• SubjectSchool planning
• Learning strandfundamental operating principles of AI / human-AI communication and collaboration / reflection on the application of AI
• Official documentsBlueprint for Digital Education Development in Primary and Secondary Schools · AI Literacy Learning Framework for Primary and Secondary Schools · Guide to Using AI in Teaching in Primary and Secondary Schools · EDB Circular No. 156/2026 · EDB Circular No. 107/2026
• Time neededAbout 90 minutes for the first fill-in; then one review a term, about 45 minutes each

An eight-column risk register that turns ten school AI risks into one line each, with a post holder and a next review date; plus a three-step incident process, handling directions for four incident types, and a one-page incident record you can photocopy on its own.

• AudienceTeachers
• ThemeSchool planning

• EquipmentEditable Word register; projector or print; the existing-controls and residual-risk columns are best filled in with the school's digital education task group present
• Version1.0
• Last updated2026-10-30

A good risk register is not a record written afterwards. It is the answer already written down on the day something goes wrong: what happens first, which post does it, and where it stops.

What this solves

This is the year schools started really using AI: teachers use it to prepare lessons and set questions, pupils use it for coursework, admin colleagues use it to tidy documents. The more it is used, the more the risks scatter — a set of usage guidelines here, a data-handling note there, a purchasing memo, a few emails. When an incident actually happens, three things usually go wrong:

• Guidelines but no register. The guidance says “mind the privacy”, but nobody can say which risk, which post is following it up, and when it will next be looked at, so the same arguments come round every year.
• A register exists, but nobody knows what to do first. The staff group chat fills with “what now” while pupils’ names and screenshots go out along with it.
• An incident becomes a person’s fault. A mis-sent file, a mistaken use, turns into a judgement about one teacher, and the next incident goes unreported.

This page does two things only: an eight-column risk register, and a three-step incident process. It does not design teaching, and it does not set the technical specification for buying a tool.

This page does not replace the school’s data-breach, complaint or personnel procedures, and it does not give legal advice. Anything that must be reported outside the school is handled under the school’s existing procedures and the requirements of the law; this page names no authority, no number of days and no deadline, because those differ between schools and change with the law. A figure printed here would be wrong for some readers.

Purpose

By the end of the register a school will be able to:

1. Write all ten school AI risks as one line each on one eight-column register: a description, a setting, a post holder and a next review date.
2. Know on the day of an incident that the first step is to contain, without waiting for a meeting to decide who is to blame.
3. Leave a one-page record of what happened, what was done that day and what follows — and leave no label on a pupil.
4. Tell the four incident types apart (data, content, use, operations) and know the handling direction for each, instead of starting from nothing every time.
5. Join up with the documents the school already has: the staff data-handling note, the minimum-data checklist, the responsibility table, the incident record form. Each does its own part, and none repeats another.

The register is built on the values of nurture first, literacy first, application-oriented and AI for good, and on the three learning strands of AI literacy (the fundamental operating principles of AI, human-AI communication and collaboration, and reflection on the application of AI). Risk management is not a way of keeping AI out of the classroom: it is the layer that lets learning and teaching use a tool with confidence. Once the risks are written as checkable fields, a teacher can let pupils try things in a lesson; and the values of critical discernment, timely and appropriate use, human-AI collaboration and technology for good finally have concrete settings to be taught in.

One principle governs the fill-in: the register records the school’s risks, not a colleague’s risks. That is why column seven carries a post, never a name.

How to use it

The file is a Word register in five parts: purpose and boundaries, the eight-column register (ten risks listed, waiting to be filled), the three-step incident process with the four incident types, the one-page incident record, and sign-off with review. The digital education task group should convene it; the first fill-in takes about 90 minutes.

Step 1: eight columns, and no ninth

The register has eight fixed columns. The more columns there are, the fewer get filled, so a need that does not fit goes in the row’s note rather than into a new column.

Column six has three values on purpose. Three values can be judged in the room and checked at a staff meeting; a score turns the discussion into an argument about whose number is right. The whole register carries no score, no percentage and no total.

Step 2: ten school risks, and only ten

Ten is a starting set, not a limit on what a school may care about. Each is one line and has no sub-items. A school that needs an eleventh adds it after the tenth and renumbers, rather than splitting one of these into parts.

1. Pupils entering personal data — a pupil types a name, class, address, photograph or family detail into a tool, when the task never needed it.
2. A teacher using an unapproved tool — to keep up with the timetable, a teacher uses a tool the school has not approved, and nobody has checked its accounts, terms or where the data goes.
3. Wrong content taught as fact — the tool’s answer is wrong, but it is complete and confident, so it reaches the classroom as fact.
4. AI-generated images of pupils’ faces — a pupil’s photograph or facial features are uploaded, generated or edited, with no handling of consent or retention.
5. A tool doing the coursework, so the marking is wrong — the work was produced by a tool, the teacher marks what was handed in, and what was assessed is not the pupil’s ability.
6. Terms changing, or a sudden charge — the tool changes its terms, adds a charge or withdraws a free allowance, and the teaching plan has to change mid-term.
7. A shared or stolen account — one account is used by several people, or is taken over, and the usage record no longer maps to a real person.
8. Copyright in generated content — a generated image, passage or piece of music is substantially similar to an existing work and is used in something published.
9. Emotional dependence or overuse — a pupil outsources thinking and emotional support to the tool, cannot start without it, or its use eats into sleep.
10. The school’s reputation — any of the above handled badly, and then discussed in public in a way the school did not expect.

Of the ten, items 1, 4, 7 and 8 are about data and access; 3 and 8 about content; 5 and 9 about use; 6 and 10 about operations. That mapping is what step 4 uses.

Step 3: three steps, and only three

On the day of an incident, do three things in order. Within all three, no pupil involved is discussed in a group chat — not a name, not a class, not a screenshot, not a transcript.

The commonest failure is between steps 1 and 2: someone is in a hurry to “sort it out”, deletes the records before the assessment, and step 2 can no longer establish the extent. Preserve first, handle second. Anything reported outside the school in step 3 goes through the school’s existing procedures and the requirements of the law; this page names no authority, no number of days and no deadline.

Step 4: four incident types, two sentences each

Four types, no more. To classify, ask one question: is this first a harm to data, to content, to use, or to operations?

• Data — an incident involving personal data, an account or access rights: stop the data leaving first, then establish who obtained what. The record says which kind of data and how many people were involved, not pupils’ names.
• Content — an incident involving wrong, false, deepfaked or inappropriate content: stop that content being used or forwarded first, then find which lessons and materials it reached. The point is correction and re-teaching, not tracing who forwarded it first.
• Use — an incident involving a tool doing the work, over-reliance or misuse: go back to the design of the task and the assessment first, then decide on a redo or an adjusted assessment. The point is the pupil’s learning, not a label on a pupil or a teacher.
• Operations — an incident involving terms, charges, an interruption of service or reputation: establish what the school can still rely on first, then agree one line to say outside. The point is that teaching does not stop.

Step 5: the one-page incident record

One page, photocopiable on its own, with fixed fields: date and time; the post of the person who found it; the incident type (one of the four); the event in one sentence; what was done to contain it that day; who and what was affected; whether it must go down the established route; the follow-up and the post responsible; the date closed.

The record holds facts and handling only, never a judgement. No words like negligent, careless or a breach of the rules, and no pupil’s name. This is the only record format on this page. If the school already has an incident record and reporting form, use that form and add these fields to it rather than keeping a second set.

Step 6: who maintains it, and how often it is reviewed

The digital education task group maintains the register and reviews it once a term: check the dates in column eight line by line, update the residual risk in column six, and mark each finished row as still live or closed. The review feeds the self-evaluation cycle of the School Development Plan and/or the annual school plan, so it points at the same objectives as the school’s other documents.

Where this page stops, and the other documents start (each does its own part, with no overlap):

• The staff data-handling note already lists what may and may not be entered; this page does not repeat the list, it cites it as an existing control in column five.
• The minimum-data checklist is the step before anything is entered; this page does not rewrite it.
• Responding to AI-generated harassment is casework and personnel handling; this page lists it only as an incident type.
• The responsibility table allocates responsibility; column seven here carries a post.
• The incident record and reporting template is the single incident record form; this page sets only the three steps and the one-page format, and adds no second form.
• The annual AI review pack is the yearly self-evaluation and evidence file; this page is the day-to-day register and response, and neither replaces the other.
• The three pupil-facing exercises (bias lab repair, deepfake response game, safe-upload rehearsal) are things pupils do; this page does not repeat them.

Review checklist

Before the register goes back, check each item:

• The register has eight columns and no ninth has been added.
• All ten risks are present, one line each, with no sub-items.
• Column six holds only low, medium or high; there is no score, no percentage and no total anywhere.
• Every entry in column seven is a post; no name appears anywhere in the document.
• Every row in column eight has a real date.
• All three steps are present, and no record was deleted before step 2 was complete.
• All four incident types are present and no fifth has been added.
• The rule that no pupil involved is discussed in a group chat is written into the document.
• Nothing names an authority, a number of days or a deadline for reporting outside the school; the wording is the school’s existing procedures and the requirements of the law.
• The page states that it does not replace the school’s data-breach, complaint or personnel procedures and gives no legal advice.
• No commercial tool is named, and no supplier or account name appears.
• The next review date and the post responsible for it are filled in.

Files

Whole-school AI risk register and incident response (editable Word document)

Five parts: purpose and boundaries; the eight-column register with the ten risks listed and the residual risk, post and next review date left blank; the three-step process with the four incident types; a one-page incident record that photocopies on its own; sign-off and review. The counts are fixed inside the document - eight columns, ten risks, three steps, four types - so a school fills in the content without changing the structure.

Official documents

Every resource in this library is written to the AI Literacy Learning Framework for Primary and Secondary Schools and falls within the policy scope of the Blueprint for Digital Education Development in Primary and Secondary Schools.

This resource maps to:

• Blueprint for Digital Education Development in Primary and Secondary Schools (three-fold leaflet, p.2): overall objective 3, enhancing infrastructure and resource provision, asks for a digital education environment that is safe and efficient. This register turns “safe” into eight checkable columns and three steps, so the objective has something to be followed up on inside the school.
• Supplement I: AI Literacy Learning Framework for Primary and Secondary Schools (p.1-4): sets out the three learning strands and progression across four learning stages. Every risk in the register should be able to say which strand’s learning and teaching it affects, or it is hard to prioritise.
• Supplement II: Guide to Using AI in Teaching in Primary and Secondary Schools (pp.2-2 to 2-5): sets requirements for a school’s use of AI tools, including starting from a teaching need, handling learning data, and the teacher’s professional gatekeeping in the classroom. The existing-controls column is written against it.
• EDB Circular No. 156/2026 (Appendix 1, section 2): builds digital and AI education into the School Development Plan, establishes a school-based implementation timetable and uses the PIE self-evaluation cycle; the same section’s “developing a smart campus” covers using AI for school administration to improve governance and efficiency, which is where most of risks six to ten come from. (Appendix 1, section 4) cultivates critical discernment and the ability to identify error and false information, including the risk of deepfakes, and requires compliance and safety: protecting personal privacy, respecting intellectual property, upholding academic integrity, and never using AI for cyberbullying or to generate inappropriate material. (Appendix 3, basic unit 2) covers AI ethics principles, defining roles and permissions, human oversight and decision responsibility, data privacy and security, and identifying risk and the direction of response - which is where the three steps and the post in column seven come from. (Appendix 3, advanced unit 2) covers using an ethical framework to set school-based guidelines, an approval process and a code of AI use.
• EDB Circular No. 107/2026 (paragraph 4): the self-evaluation cycle, and building digital education into the School Development Plan and/or the annual school plan; the termly review of this register joins that same cycle.

Learning strand: fundamental operating principles of AI / human-AI communication and collaboration / reflection on the application of AI

Official documents page: https://www.edb.gov.hk/en/edu-system/primary-secondary/applicable-to-primary-secondary/it-in-edu/debp.html

This page is an independent Beets teaching resource, not an official EDB document. The eight columns, the ten risks, the three steps and the four incident types are a suggestion, and a school may adapt them. This page does not replace the school’s data-breach, complaint or personnel procedures, and it does not give legal advice.

Feedback

If you have any comments or questions about this resource, or an idea for the next one, tell us on the feedback page.