Stellenbosch rejects blanket rules for AI use in university assessments

The South African university’s new guidance allows generative AI to be prohibited, restricted, permitted or required according to the learning outcomes and disciplinary context of each task

Stellenbosch University’s new resource allows academics to set different conditions for generative AI use according to the learning outcomes and context of each assessment

Stellenbosch University has placed decisions about generative AI use at the level of individual modules and assessments, rejecting a single institution-wide rule that would either prohibit or permit the technology across the university.

Its new AI in Teaching, Learning and Assessment resource gives academics guidance on setting their own conditions for AI use. Depending on the purpose of an assignment, lecturers may prohibit, restrict, allow, encourage or require students to use AI.

The South African university’s position makes intended learning outcomes the deciding factor. Academics are expected to consider what a task is designed to assess, what evidence would demonstrate that learning has taken place and whether AI would support or undermine that process.

Students remain responsible for the accuracy, originality and academic standard of their work, even when AI use is permitted. They may also be required to explain or defend a submission where a lecturer needs further evidence that the stated learning outcomes have been met.

Liezl van Dyk, Deputy Vice-Chancellor: Academic at Stellenbosch University, announced the website on LinkedIn. She described it as providing “practical, evidence-informed guidance for academics, from quick-reference resources to more in-depth material on AI in teaching, learning, assessment, research and scholarship.”

Developed through the Division for Learning and Teaching Enhancement, the resource is aligned with Stellenbosch University’s Vision 2040.

Lecturers must define the boundaries

Stellenbosch does not currently have a single university-wide AI policy. It instead uses an ethical position statement and guidelines organized around authenticity, fairness, accountability and transparency.

The four principles are presented as tools for making academic decisions, rather than a checklist for enforcing compliance. This allows the university to maintain a common institutional position while recognizing that acceptable AI use will differ between disciplines and forms of assessment.

A programming task, laboratory report, literature review, design project or professional placement may each require a different approach. An AI tool that supports learning in one setting could remove the intellectual work being assessed in another.

Lecturers must therefore tell students whether AI can be used, what forms of assistance are permitted, whether that use must be declared and what responsibility remains with the student. Stellenbosch states that academics should also explain the educational reason for the conditions, rather than communicating only whether AI is allowed.

Its assessment guidance identifies three broad approaches. Academics may allow or encourage responsible AI use, prohibit it where doing so serves a pedagogical purpose, or require students to use it as part of a task.

Where a particular AI platform is compulsory, an alternative should be available for students who do not want to register for that service. This introduces an access consideration alongside the academic judgment about whether AI belongs in the assessment.

The university does not position prohibition as its default response. It also warns against relying solely on AI detection systems when investigating suspected misuse. Where the authenticity of a submission is questioned, lecturers can seek other evidence, including discussing the student’s working process or conducting an additional oral assessment.

Stellenbosch’s Senate-approved position statement on the ethical use of AI took effect on January 1, 2025. It holds students and staff accountable for the integrity and accuracy of AI-assisted work and states that generative AI tools cannot be credited as authors of scholarly work.

Substantive AI contributions must be declared

The guidance distinguishes between limited technical assistance and AI use that makes a substantive contribution to academic work.

Basic language, editing or coding support may not always require disclosure unless a lecturer has specifically requested it. Generative AI used to produce material such as text, images, music or code should be declared. Where the distinction is uncertain, the university advises students to disclose the use.

For postgraduate research, students are expected to discuss their proposed use of AI with their supervisors before submitting a thesis, dissertation or research assignment.

A dedicated declaration records whether AI contributed to activities including idea generation, literature searches, research methods, experiment design, data analysis, code development, editing, writing or citation formatting. Students must identify the tool, explain its purpose, describe the extent of its contribution and justify why it was used.

The university presents the declaration as a basis for discussion and record-keeping, rather than an automatic disciplinary mechanism. Ethical and transparent AI use that is consistent with supervisory guidance and disciplinary expectations is not, by itself, grounds for a penalty.

Responsibility for the final submission cannot be transferred to the technology. Postgraduate students must still account for its accuracy, originality and integrity.

AI literacy framed as academic judgment

Stellenbosch defines AI literacy more broadly than learning how to operate individual tools. Its guidance covers evaluating outputs, identifying bias and limitations, checking information, understanding ethical implications and recognizing when AI use could interfere with learning.

There is no mandatory university-wide AI literacy certification for academic staff. Support is provided through institutional guidance, workshops, faculty discussions, short courses and practical resources, allowing faculties and disciplines to develop approaches suited to their own teaching and knowledge practices.

Two self-paced Digital Education Council courses are available during 2026. AI Literacy for All takes approximately four hours, while the Certificate in AI for Higher Education requires around 20 hours and is intended for lecturers, administrators, institutional leaders and students.

Registration requires a Stellenbosch University email address. The university says access should be provided within 24 to 48 business hours, after which participants have 180 days to complete the selected course.

The resource also addresses equity, inclusion and decolonization. AI could widen access to explanations, feedback and multilingual assistance, but the university notes that systems may reproduce existing biases or underrepresent particular languages, regions and knowledge traditions.

Several AI-related initiatives, including AI tutoring systems, are being explored across Stellenbosch.

Its guidance ultimately treats the central issue as one of evidence. As generative AI becomes capable of producing increasingly polished academic work, Stellenbosch is asking lecturers to determine whether a submission still provides a valid representation of what the student knows, understands and can do.