What good school AI policies have in common

South Africa has no national guideline on AI in schools yet. When we checked again on 12 October 2026, we found none from the Department of Basic Education and nothing on AI in assessment from Umalusi, the IEB or SACAI. The Western Cape Education Department's eLearning Directorate compiled a position paper in November 2025, but it is not departmental policy. So each school writes its own.

Before we built our AI policy wizard, we read international guidance (UNESCO, the OECD, UNICEF and education departments in six countries), the South African law that applies to schools, the exam bodies' rules, the AI providers' age limits, the research on detectors and AI policies published by real schools here and abroad. Here is what we found.

What the best policies share

The clearest policies say plainly what is allowed, for whom, and who decides. They share eight things.

1. Rules that step up with age. Singapore's Ministry of Education is a good example: no work that requires AI in Primary 1 to 3, then from Primary 4, supervised use of AI tools designed for education, and declared use on set tasks in secondary school. Many policies skip this, yet most general AI chatbots we checked set a minimum age of 13, and Claude sets 18.
2. One list of approved tools, and one person who approves new ones. Before a tool goes on the list, someone checks what it does with personal information, where it is stored and what ages it allows.
3. No personal information in tools the school has not approved. Ireland's guidance puts it well: treat anything you type into an AI tool as public unless its terms say otherwise.
4. A person checks every AI output and stays responsible for it. A report comment, a worksheet or a letter home is the teacher's work, whoever drafted it.
5. Openness about AI use. Learners say when and how they used AI. Cambridge and the UK's Joint Council for Qualifications ask learners to name the tool and the date and keep a record of their prompts.
6. A level of AI use for each task. Instead of one rule for all work, the teacher sets a level for each task, from no AI at all to AI used throughout. The AI Assessment Scale by Mike Perkins and colleagues is one such scale. When AI must be kept out, the work is done under supervision.
7. Safeguarding that names deepfakes. England's safeguarding guidance for 2026 treats a nude image made or altered with AI like any other nude image, and good school policies say how staff respond and who they report to.
8. A short review cycle. Usually once a year, and sooner when the tools or the law change.

What South African schools must add

A policy written overseas cites laws that do not apply here.

• POPIA, especially for children. POPIA forbids processing a child's personal information unless section 35 allows it, for example with a parent's or guardian's consent (sections 34 and 35). So no learner names, marks, photos or health notes go into an AI tool unless the school has approved it, has a lawful ground and, where the provider processes the information for the school, has the written contract section 21 requires.
• Information sent abroad. Section 72 of POPIA sets the conditions for sending personal information across the border, and sending children's information to a country without adequate protection may first need the Information Regulator's authorisation (section 57).
• People make the decisions. POPIA section 71 protects people from decisions with legal or substantial effects that rest solely on automated processing. Our wizard goes further: a person decides marks, admissions, discipline and staff matters.
• Your exam body's rules. NSC schools follow the Department of Basic Education's rules on authentic work in SBA and PATs. Only two 2026 Grade 12 PATs mention AI: CAT shows how to acknowledge ChatGPT, and Visual Arts allows AI images as inspiration only. Other subjects need the school's own rule. Cambridge schools follow Cambridge's policy on AI in coursework. IEB and SACAI schools should ask their exam body: we found nothing public on 12 October 2026.
• The law on fake images. An AI sexual image of a learner is not a prank. The Sexual Offences Act covers sexual images of children 'however created', whether the person shown is 'real or simulated', and the Cybercrimes Act covers intimate images that are 'real or simulated, and made by any means'. Anyone who knows of a sexual offence against a child must report it to the police at once (Sexual Offences Act section 54), and the Children's Act requires teachers who conclude on reasonable grounds that a child has been sexually abused to report it. In October 2025 parents at a Western Cape high school opened police cases over AI nude images of girls, with no evidence they were made in school hours. A policy must also cover fake images made at home.
• The right people approve it. In a public school the governing body adopts the policy, and consequences for learners belong in the code of conduct, which the Schools Act says the governing body adopts after consulting learners, parents and educators.

Mistakes to avoid

• Banning everything. A ban nobody can check moves AI use out of sight and leaves learners without guidance. California's state guidance steers schools towards clear rules on what is allowed instead.
• Trusting AI detectors. Detectors get it wrong, and a 2023 study in the journal Patterns found seven detectors labelled more than half of a set of essays by non-native English writers as AI-generated. The test was in English, but many South African learners write in a language that is not their home language, so the risk is real here. England's exams regulator, Ofqual, advises starting with a conversation about how the learner did the work.
• Copying another country's policy. A UK template cites UK law and exam rules, but a South African school answers to POPIA, the Children's Act and its own exam body.
• A 'No AI' rule for homework. Nobody can check work done at home. The researchers behind the AI Assessment Scale warn against 'No AI' labels that cannot be enforced: if work must be done without AI, do it in class.
• Forgetting to check. In April 2026 Minister Solly Malatsi withdrew the draft national AI policy because its reference list included sources that do not exist, and Cabinet approved the withdrawal in June. The same goes for every AI draft in a school, a policy included.

Write your school's policy

Our free AI policy wizard puts this to work. Answer a few questions and it drafts a policy with the sources for every section, as a Word document, plus a one-page summary for parents. It runs in your browser, so nothing about your school reaches us unless you choose to send it.

It is a well-sourced starting point, not legal advice. Have a lawyer or POPIA specialist check it, then ask your governing body to adopt it, and review it once a year, because AI tools change fast. If you would like help, we work through AI policies with leadership teams as part of our consulting, and we run AI talks for staff, parents and learners.