Singapore did not ban AI. It split assessment in two: what students can do alone, and what they can do with AI, out loud.
The short answer: Singapore's Ministry of Education keeps national and school-based exams proctored with no AI allowed, while permitting AI in some coursework if students declare how they used it. This works only if teachers can design tasks that reveal student thinking, a skill set the UNESCO AI Competency Framework for Teachers treats as core, so assessment redesign depends on teacher readiness, not detection software.
On 7 October 2026, MOE answered a parliamentary question from Jackson Lam, MP for Nee Soon GRC, who asked whether MOE would develop ways to check that students can still write, research and solve problems without AI. The reply, also reported by Little Big Red Dot on 13 October, was plain. Exams stay proctored. AI stays out. There will be no separate test of AI-free ability. Where a task allows AI, students "must state how AI is used and they remain responsible for their own work." And MOE will keep reviewing its assessment approaches "as AI capabilities evolve."
Read it quickly and it sounds like a policy statement. Read it slowly and it is a design brief for every school that has to live under it.
What exactly did Singapore's MOE decide about AI in exams?
Three things. First, national and school-based examinations are typically proctored and disallow AI. Second, in coursework such as some research assignments, AI use can be allowed when it suits the task, but students must disclose it. Third, MOE has set no date and committed to no specific change, only to continued review.
Notice what is missing. There is no detection mandate. There is no approved-tool list in this reply. The declaration rule does not say how students declare, in what format, or what happens if they do not. That detail sits with schools.
Why does an AI declaration rule put the weight on teachers?
A declaration is only as good as the task it is attached to. If a research assignment can be completed by pasting a prompt into a chatbot, a line saying "I used AI for the draft" tells a teacher nothing about what the student understands. If the task asks students to defend a choice, compare two sources the teacher supplied, or explain a step in their own words, the declaration becomes useful evidence.
That is assessment design, and it is a trained skill. MOE's wording says assessments are designed so students still show their own thinking, judgement and understanding. Somebody has to do that designing, for every subject, every term.
Detection is a purchase. Redesign is a capability. A scanner buys one exam season. A teacher who can build AI-resilient tasks compounds across every class they ever teach.
Is this a model for schools in Manila, Kuala Lumpur or Mumbai?
Partly. The two-track logic travels well: protect a few high-stakes, proctored moments where independent mastery is measured, and open up the rest with transparency. Exam-centric systems already have the proctored half. What most schools lack is the second half: the coursework tasks and teacher skill to make disclosure meaningful.
It would be a mistake to copy the headline and skip the infrastructure. Singapore has been investing in teacher time and AI training alongside its policy, as we covered in its 2026 plan to cut teacher workload with AI. A school with large classes and thin PD budgets cannot declare its way to integrity. It has to decide where to put limited teacher hours, and the evidence points to assessment redesign over surveillance.
What should school leaders do about AI declarations and exams?
Sort your assessments into two columns. List what must be proven independently and what can be done with AI. Be explicit with students and parents about which is which, task by task.
Write a declaration format before you write a penalty. A one-paragraph template (tool used, what it was asked to do, what the student changed and why) is more useful than a signature line. Keep it short enough that students will actually complete it.
Redesign one high-stakes coursework task per department this term. Add an oral defence, a live in-class step, or a process log. Not next year. This term.
Train teachers on assessment design, not just tool use. The skill that makes the declaration meaningful is task design. Map your PD against the UNESCO AI Competency Framework for Teachers and check where assessment sits.
Set a review date. MOE itself is reviewing as AI evolves. Your school policy should have a date, too.
Frequently asked questions
Does Singapore allow AI in national exams?
No. According to MOE's 7 October 2026 parliamentary reply, national and school-based examinations are typically proctored and disallow the use of AI. This lets MOE assess what students can do independently.
Do Singapore students have to declare AI use?
In coursework where AI use is allowed, yes. MOE says students must state how AI is used and remain responsible for their own work. The reply does not specify the declaration format or penalties.
Is MOE planning a separate test for AI-free skills?
According to the reporting, no. MOE keeps exams proctored and AI-free and says it will continue reviewing its assessment approaches as AI capabilities evolve, without committing to a date or specific change.
How can schools in other countries apply Singapore's approach?
Separate assessments into independent and AI-permitted tasks, require simple declarations, and redesign coursework so it shows student thinking. The limiting factor is usually teacher training in assessment design, not policy.
thegurucool builds teachers' AI competency through structured professional development for teachers and educators in schools, colleges and institutions. The TEACH-AI framework covers 7 domains, from first exposure to confident AI-era assessment design.
Related reading: Singapore's 2026 Plan to Cut Teacher Workload With AI: What Changed · Mandatory AI Literacy Is Here. Teacher Training Still Assumes a Computer Lab.