Explains institutional arrangements in relation to AI-supported assessment, covering scope, evidence, decision authority, material exceptions and continuing assurance.
Against the background of the rapid adoption of generative AI tools, education authorities and providers should review how institutional arrangements for AI-supported assessment is defined, implemented and evidenced. The central issue is the meaning of the expectation in practice, including its scope, the evidence needed to demonstrate it and the circumstances in which it may not apply.
Meaning in practice
The public-interest assessment of AI-supported assessment should consider access, learning, fair treatment and the reliability of information on which learners make consequential decisions. Verified fact, policy expectation and discretionary institutional choice should remain distinct in the record. Later review should not obscure whether the earlier position rested on fact, policy or judgement.
The system and institutional dimensions of the applicable expectation should be considered together. In the context of AI-supported assessment, assessment should provide valid and sufficiently consistent evidence that the stated learning outcomes have been achieved by the learner receiving the result. Governance of assessing institutional arrangements for the applicable requirement requires a clear allocation of authority, information and follow-through. Escalation should place material evidence before the authority capable of an effective response. Responsibility for the matter should be identifiable at each consequential decision point. Delegation should identify both the operating role and the body retaining oversight of learner impact.
Analysis should make its decision rule explicit. In the context of AI-supported assessment, the subject should be examined as a connected system of policy, people, resources, decisions and evidence. The assessment should follow authority and information across functional boundaries and verify completion of required action. The method should prevent an unfavourable result from being dismissed through an unrecorded change in interpretation.
Governance of the assurance conclusion requires a clear allocation of authority, information and follow-through. When examining AI-supported assessment, material matters should be referred to the body authorised to act or accept residual risk. Conformity should not be inferred from a policy document alone; operating records and outcomes should show that the stated arrangements are in use. Terms governing eligibility, support, assessment, reporting or review should prevent materially different treatment without recorded justification.
Risk assessment should give particular attention to tasks that do not assess the stated outcome, weak assurance of authorship or performance, and reasonable adjustment altering the assessed outcome. A provider should also consider inconsistent judgement between markers or locations and results used beyond the evidence they support. Review of the applicable expectation should include the experience of affected learners, particularly where aggregate reporting may conceal exclusion, delay or unequal treatment.
For AI-supported assessment, readily available material should not define the enquiry if it cannot answer the relevant decision question. Examination of the matter should give particular attention to adverse cases, unequal effects and errors that learners may be unable to identify or remedy after the event.
Responsibilities and material risks
Arrangements for AI-supported assessment should provide accurate information, timely support and an accessible route for correction or review without adverse treatment. A competent review of the applicable expectation should map the complete process, identify the intended result and responsible authority at each stage, and test normal cases together with exceptions. The principal risks associated with the control should be assessed as connected conditions.
Interpretation of AI-supported assessment should produce a test that another competent reviewer can apply to comparable evidence.
Risk assessment for the assurance conclusion should consider severity, reach, duration, recurrence and detectability, with escalation where learner impact may be material. The review method for the applicable requirement should connect the question under examination to suitable evidence and a conclusion no broader than the tested scope. Within the scope under review, validity without adequate consistency may expose learners to unequal judgement. For AI-supported assessment, the volume of documentation is not a measure of conformity. Relevance, integrity and coverage are more important than the number of records produced. No exception should continue without a documented basis, accountable approval and scheduled review.
Responsibility for assessing the applicable requirement for the control should be identifiable at each consequential decision point. Accountability for learner impact should remain explicit when delivery tasks are delegated. When examining AI-supported assessment, this enables later review to separate substantive change from correction, reclassification or expanded coverage. A superseded conclusion should be retained where it formed the basis of a material decision.
Public reporting on the assurance conclusion should distinguish established fact, analytical judgement and planned action. Reporting on the assurance conclusion should distinguish established fact, analytical judgement and planned action. For decisions concerning AI-supported assessment, material revisions should retain their reason and effective date.
Progress on assessing institutional arrangements for AI-supported assessment under review is not the amount of policy or documentation produced.