A school AI policy should not begin and end with whether students may use a chatbot. AI already affects search, writing, images, recommendations, assessment, administration and communication. Governance must protect children and professional judgement while leaving room for responsible learning.
This is an implementation framework, not legal advice. Schools should align it with Ghanaian law, regulator guidance, institutional policy and platform terms, with qualified review where required.
1. Purpose and principles
State what the policy is trying to protect and enable: learning, human agency, child rights, fairness, privacy, academic integrity, accessibility and professional responsibility.
2. Scope and definitions
Define who and what is covered: learners, staff, contractors, approved platforms, generative systems, automated decision tools and AI features embedded in existing software.
3. Permitted, conditional and prohibited uses
Create clear examples by role and task. Brainstorming may be permitted with disclosure; processing identifiable student data may be prohibited; high-stakes decisions should require accountable human review.
4. Data protection and privacy
Specify which information must never be entered, how accounts are created, what consent is needed, where data may travel, how long it is retained and who assesses platform terms. Treat student work, images, health, behaviour and assessment data as sensitive.
5. Child safety and well-being
Address age appropriateness, harmful content, manipulation, dependency, impersonation, sexual abuse material, harassment and incident escalation. UNICEF’s child-centred AI guidance makes children’s best interests, safety, inclusion and development central requirements.
6. Assessment and academic integrity
For each assessment type, state what assistance is allowed, what must be disclosed and what evidence of process is required. Use oral defence, classroom checkpoints, drafts, citations and demonstrations where appropriate.
7. Accuracy, bias and human review
Require users to verify consequential claims and identify likely bias or missing context. AI output should not be treated as an authoritative source. High-impact decisions about learners need named human accountability.
8. Transparency and authorship
Define how staff and learners acknowledge AI assistance and how the school communicates automated uses to families. Transparency should be simple enough to practise consistently.
9. Procurement and vendor review
Before approval, examine learning purpose, age terms, accessibility, data handling, model behaviour, security, moderation, export and deletion, support, total cost and exit options. A popular tool is not automatically an appropriate school tool.
10. Training, incidents and revision
Name the policy owner, staff training requirement, reporting route, response process and review schedule. Record lessons from incidents and classroom trials. Review at least annually and after material platform or regulatory change.
Policy is a practice
Students, teachers, leaders and families should understand not only the rules but the reasons. Use scenarios during induction and revisit them as tools change. A policy hidden in a handbook cannot govern a fast-moving classroom.
Download the Ghana School AI Policy Template and Implementation Guide.
A policy should govern judgement, not chase every new product
An AI policy written around today’s tool names will age quickly. A durable policy governs categories of decision: what information may enter a system, when human review is mandatory, how assistance is disclosed, which uses undermine assessment and who is accountable when harm occurs. Products will change; these institutional responsibilities will not disappear.
The policy should be understandable at the point of action. A teacher planning tomorrow’s assessment and a learner completing tonight’s assignment need examples, not only principles. Convert each principle into realistic permitted, conditional and prohibited scenarios.
The assessment question is deeper than cheating
AI exposes a weakness that already existed: many assignments reward a finished response while revealing little about the learner’s thinking. Schools should redesign assessment to make process visible through drafts, oral defence, in-class application, source checking, reflection and transfer tasks. Disclosure matters, but disclosure alone does not prove learning.
The policy should distinguish assistance that supports learning from substitution that removes the capability being assessed. The same use may be appropriate in one task and invalid in another because the learning purpose differs.
Policy must reach procurement
Responsible classroom behaviour cannot compensate for an unsafe institutional purchase. Before approval, schools should examine age requirements, data use, retention, model training, security, accessibility, advertising, moderation, account control, deletion and exit arrangements. Someone must be authorised to say no when the educational benefit does not justify the exposure.
Test the policy with four incidents
- A teacher enters identifiable learner work into a public AI service.
- A learner submits an accurate answer they cannot explain.
- An AI-generated image humiliates another learner.
- A school uses an automated recommendation in a consequential decision.
If leaders cannot explain the response, evidence, escalation route and accountable person for each case, the policy is not yet operational. Good governance should make responsible action easier before an incident, not merely assign blame afterward.
Implementation is part of the policy
Publish a short learner guide, a staff decision tree and an assessment disclosure format alongside the full policy. Train leaders to receive incidents without panic, and review anonymised cases each term to identify where guidance remains ambiguous. A policy no one rehearses will be interpreted inconsistently under pressure.
Schools should also create a review trigger for material changes: a new product, a major platform update, a regulatory change or evidence of harm. Annual review is the minimum rhythm, not the only moment when governance can learn.
Digital safety conversations must begin with children’s real curiosity
During Metaverse Safety Week, I described what happens when young people first encounter immersive technology: “They get so excited, they get so curious, they come up with so many questions.” That curiosity is precisely why governance cannot be reduced to restriction. Schools need structures that help young people explore while learning to recognise manipulation, privacy risk, unreliable output and unequal power.
The 2024 Youth Empowerment Community Lab report noted that my perspective was grounded in introducing advanced technology where basic access can still be a luxury. That contrast matters for AI policy. Ghanaian schools must address sophisticated risks without assuming sophisticated infrastructure, specialist staff or universal personal devices.
A good policy therefore has two obligations: protect children from foreseeable harm and protect their right to meaningful technological participation. Safety without opportunity reproduces exclusion. Opportunity without safeguards transfers unacceptable risk to learners and teachers.
