AI Assessment Scale for African Classrooms: A Practical Guide

Last reviewed: October 2026.

Generative AI has made one question unavoidable for every school and assessment body: when a learner submits work, what did they actually do? Banning AI does not answer it, and detection tools cannot answer it reliably. What works is being clear, task by task, about how much AI use is allowed and designing assessment so that the evidence still means something.

This guide shows how the AI Assessment Scale can help African schools and assessment teams do that, how it fits national assessment rules, and how assessment professionals can use AI to design better assessments without giving up their own judgement. It builds on my earlier piece, Adapting the AI Assessment Scale for South African schools, which remains the detailed case study.

Why assessment is where the pressure lands

Most conversations I have with school leaders about AI still start with cheating. That is understandable, but it frames AI as a discipline problem when it is really a design problem. If a take-home essay can be produced by a chatbot in seconds, the task no longer tells a teacher much about what the learner knows. The useful response is to decide what each task is meant to show and set the AI rules to match.

International guidance points the same way. UNESCO’s Guidance for generative AI in education and research, published in September 2023, calls on institutions to validate AI tools for ethical and pedagogical appropriateness and urges a rethink of assessment rather than a reliance on policing.^[1]

The AI Assessment Scale

The AI Assessment Scale (AIAS) was developed by Mike Perkins, Leon Furze, Jasper Roe and Jason MacVaugh. Its first formal version was published in the Journal of University Teaching and Learning Practice in April 2024.^[2] Rather than treating AI as either permitted or banned, it gives teachers five levels for describing how AI may be used in a task.

The authors have since refined it. An updated version by Perkins, Roe and Furze, published in the same journal in 2025, reframes the scale as a tool for redesigning assessment to protect validity, presents the levels as non-hierarchical, and adds a new level for exploring AI itself.^[3] The current levels, with resources and translations, are on aiassessmentscale.com:^[4]

Level 1, No AI: the task is completed without AI, typically in a supervised setting.
Level 2, AI Planning: AI may support brainstorming, planning and research, but the learner develops the work.
Level 3, AI Collaboration: AI may help draft and refine, with the learner evaluating and modifying what it produces.
Level 4, Full AI: AI may be used throughout, with the learner directing the work and responsible for the outcome.
Level 5, AI Exploration: AI is used creatively or critically as part of the learning itself.

My South African case study uses the original 2024 levels, which are still widely used. The principle is the same in both versions: the teacher sets the level, explains it to learners, and designs the task so that the evidence matches. The AIAS diagrams are shared under a Creative Commons licence for non-commercial educational use.

Adapting the scale for African classrooms

The scale was designed in higher education settings with reliable connectivity. Many African schools work with very different conditions, and the adaptation matters more than the framework itself. From my work with South African schools, four realities shape how it can be used:

Unequal access. Some schools have excellent facilities; many have no reliable internet. Where access is limited, the No AI and planning levels are the practical core, with occasional structured AI sessions in a computer lab rather than an expectation of constant access.
Data costs. Mobile data remains expensive. Education authorities can negotiate zero-rated access to educational AI platforms, as many did during pandemic remote learning.
Language. South Africa has eleven official languages, yet most AI systems perform best in English. Planning and editing levels can support learners working in an additional language, but teachers should be open about the limits.
Teacher training. Many teachers have had no formal training in AI. Practical professional development on using the scale is more useful than general awareness sessions.

A No AI level is not a step backwards. In classrooms where some learners have AI at home and others do not, supervised No AI tasks protect fairness. Where AI is allowed, the task should be possible with printed materials or shared devices, so that access does not decide the mark.

Fitting national assessment rules

In South Africa, school-based assessment and practical assessment tasks are governed by the National Protocol for Assessment Grades R to 12, which standardises recording, reporting and moderation within the National Curriculum Statement.^[5] The AIAS does not replace that framework. It sits inside it, as a way for teachers to state the AI conditions for each internal task, alongside the evidence and moderation requirements already in place.

Policy is moving. In May 2026 the Minister of Basic Education, Siviwe Gwarube, wrote that the department is reviewing its digital education policy framework and developing practical guidance for schools, and that AI must not make high-stakes decisions about learners without human oversight.^[6] Provincially, the Western Cape Education Department published an AI position paper for its schools in August 2026.^[7] Until national guidance is published, a school-level AIAS policy is a sensible way to give teachers and learners clarity.

By phase, a reasonable starting point is:

Foundation Phase: No AI for learner tasks. AI, if used at all, supports teachers in planning.
Intermediate Phase: mostly No AI, with teacher-led AI Exploration activities, such as checking an AI-generated text against a trusted source as a class.
Senior Phase: No AI and AI Planning for most tasks, with occasional AI Collaboration where learners have access and the task is designed for it.
FET Phase: the full range for internal tasks, set task by task, with No AI conditions kept for supervised assessments and examinations.

Using AI to design assessment

The second half of the assessment question is how teachers and assessment teams use AI themselves. Here the opportunity is real. In Designing formative assessments with spreadsheets and AI I set out a workflow that starts with a structured spreadsheet of requirements, covering curriculum, subject, outcomes and question types, and uses AI to generate aligned questions, followed by a quality assurance stage. It began as a prototype, but the principle holds: structured inputs, then generation, then human review.

The pattern that matters is generate, then judge. AI drafts quickly; experts decide what is accurate, fair, aligned and at the right level. Retrieval from your own curriculum and source documents reduces the risk of plausible but wrong content. Bias, distractor quality and curriculum alignment are checked as part of the workflow, not as an afterthought.

Training assessment teams

This is the thinking behind Designing Assessments with AI, the course I designed and deliver as the named Cambridge trainer for the Cambridge Assessment Network. It is an accredited series of three live online sessions, two hours each, for six CPD hours, and I have delivered four cohorts so far. The course grew out of a 2023 demonstration to Cambridge’s Quarterly Global Leadership meeting.

The sessions move from how the tools work and ethical considerations, through a structured workflow for generating assessment content, to risks, quality assurance and evaluation. Participants generate items from structured inputs, evaluate them, and leave with a checklist for AI-generated assessment content. The method works across capable large language models, so it does not depend on a single tool, and the focus is formative assessment with ideas that also scale to summative work.

If you lead an assessment, publishing or curriculum team, read how the course was designed, or see current dates and booking on the Cambridge Assessment Network course page. For designing wider AI training, see How to Design AI Professional Development That Changes Practice.

Getting started in a school

Agree a school position: AI use is set per task using the AIAS, and learners are told the level in writing.
Start with one subject and one term, gather feedback from teachers and learners, then expand.
Redesign one high-risk take-home task so that it shows process, for example through drafts, oral explanation or in-class completion.
Keep supervised No AI tasks for evidence that must be the learner’s own.
Train teachers on the scale and on checking AI outputs before they set AI-permitted tasks.
Review the policy each year as national guidance develops.

For the wider picture, see AI readiness for schools, AI literacy in African schools and Sovereign AI for education systems. For workshops with your team, see Speaking or get in touch.

Sources

Niall McNulty is Director of Product and Innovation, Sub-Saharan Africa, at Cambridge University Press & Assessment.