---
title: "Assessment in the AI age: Universities need structural redesign"
author: Martin Ebner
date: 2026-11-13
url: https://postgenerative.at/indications/2026-11-13/assessment-in-the-ai-age-universities-need-structural
lang: en
type: Indication
summary: "A report by OfS and Advance HE calls for structural assessment changes: verify students’ own capabilities, deliberately accommodate AI use and align tasks with learning outcomes."
tags:
  - Education
  - Higher education
  - Assessment
sources:
  - title: "Office for Students und Advance HE (11.11.2026): Making AI work for students: reflections on institutional practice – Publikationsseite"
    url: https://www.officeforstudents.org.uk/publications/making-ai-work-for-students-reflections-on-institutional-practice/
  - title: "Office for Students und Advance HE (2026): Originalbericht, insbesondere Kapitel 2, S. 16–24, und Schlussfolgerungen, S. 42–44 (PDF; Open Government Licence 3.0)"
    url: https://www.officeforstudents.org.uk/media/nf2ffugd/making-ai-work-for-students.pdf
license: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
alternate:
  de: https://postgenerativ.at/indizien/2026-11-13/pruefungen-im-ki-zeitalter-hochschulen-muessen-das-design
---

# Assessment in the AI age: Universities need structural redesign

Published on 11 November 2026, the Office for Students (OfS) and Advance HE report examines higher education’s responses to generative AI. It identifies a fundamental problem: A strong submission no longer reliably demonstrates that students possess the intended knowledge and skills if AI can perform most of the task. Additional instructions and retrospective checks alone cannot resolve this design problem. [1][2]

### How assessment should be redesigned

1. **Start with learning outcomes.** Institutions must establish which capabilities students need to demonstrate independently and where competent AI use is itself a learning outcome. An assessment is not effective simply because AI access is prevented: It must actually measure the intended capabilities.
2. **Secure evidence of independent achievement.** For appropriate learning outcomes, the report recommends controlled conditions, including in-person examinations, vivas, observed practical tasks or short discussions of submitted work. This encompasses more than traditional written examinations.
3. **Design open assessments for realistic AI access.** Where AI is available, the task must still measure the intended capability. Examples include critically evaluating AI output or improving an AI-generated draft with reasoned changes. If AI can complete a task successfully without the knowledge or skills being assessed, the task should be replaced. Rapid model development requires ongoing testing of these approaches.
4. **Make the process visible.** The report describes checkpoints and staged assessment of work as possible approaches: Alongside the final product, assess how students research, summarise and develop their findings. These stages must also provide credible evidence of achievement.
5. **Build in fairness and clarity.** Permitted AI use must be clearly communicated. Accessibility, reasonable adjustments, unequal access to AI and possible alternatives for students who do not wish to use it belong in assessment planning. Suspicion alone does not justify a lower mark; AI detectors cannot replace fair procedures. [2]

### Bath’s example: Two lanes instead of a difficult-to-police middle ground

According to the report, the University of Bath moved from a traffic-light system to a two-lane approach in September 2026, originally developed at Sydney. In the **closed lane**, assessment conditions prevent AI access. In the **open lane**, AI is permitted and its availability is assumed in task design. Both lanes must align with learning outcomes. This is a practice example, rather than a universal requirement for all institutions. [2]

### What changes for students and institutions

Students use AI for explanations, summaries and practice questions, among other purposes. The report distinguishes learning support from outsourcing the cognitive effort needed to develop capability. For future assessment, this means providing credible evidence of independent achievement and, where AI is part of the task, critically judging its outputs. Institutions therefore need to connect assessment and curriculum redesign with discipline-specific AI literacy for students and staff. [2]

**Evidence and scope:** This qualitative report incorporates 244 higher education staff and 91 students and student representatives from at least 127 institutions. Participants were self-selecting, so findings are not representative. Engagement took place between June and September 2026. The conclusions are recommendations by OfS and Advance HE, rather than new regulatory requirements. [2]