Universities around the world – and in South Africa – are having an identity crisis, and generative AI is the trigger.

More than three years after ChatGPT walked into the lecture hall, the debate has moved well past cheating. In a recent piece in New York magazine, a Columbia undergraduate boasted of using AI to write most of his essays; in The New Yorker, the critic Hua Hsu asked, “What happens after AI destroys college writing?” What does a degree certify when a chatbot can produce a competent essay in seconds?

South Africa’s universities are not watching from the touchline – far from it. Detailed responses to TechCentral from North-West University (NWU), the University of Pretoria (UP), Stellenbosch University (SU) and the University of the Witwatersrand (Wits) reveal a sector that has abandoned an earlier, defensive posture of “AI policing” for something more serious and ambitious: a holistic, even philosophical, rethink of teaching, assessment and purpose.

The clearest sign of the shift is what universities have stopped doing. Stellenbosch “discontinued the use of Turnitin’s AI text detection functionality at the end of 2025”, says Hanelie Adendorff, senior advisor at the university’s Centre for Teaching and Learning.

The tools, she says, do not work well enough to justify the harm they can cause. Independent evaluations find their accuracy collapses the moment a student lightly edits AI-generated text, and a widely cited Stanford study found that essays by non-native English speakers were wrongly flagged as AI-written 61.3% of the time, against just 5.1% for native speakers.

‘From policing to stewardship’

In a country where most students write, think and are assessed in a second or third language, that bias is a problem. Mario Landman of the Academic Centre of Excellence, which serves private education group AdvTech and its Independent Institute of Education, argues that the sector should abandon detection, describing the shift as a move “from policing to stewardship”.

Wits’s newly approved official AI policy, adopted last month, reaches the same conclusion from a different direction: rather than police, it asks staff to set “assessments that mirror real-world professional tasks and are inherently difficult for AI to complete generically”.

Beneath the institutional differences, an important consensus is emerging. All four universities have replaced the binary question “Is AI allowed?” with a more demanding one about what students should be able to do with it, and without it.

Read: Why AI gets smarter as it scales – a Wits study has a clue

UP’s guiding principle is unambiguous: “Students must think first, and use AI later,” says the university’s media liaison, Liesel Swart. Stellenbosch is refining an “AI use bar” that helps lecturers decide when AI use should be “prohibited, restricted, allowed, encouraged or required”.

NWU lets lecturers set the level of permitted AI per assessment, “ranging from no AI, to brainstorming or editing support, to partial or full use”, and the Wits policy likewise turns on “clearly defining the permissible and impermissible uses of AI within specific academic tasks”.

The second point of agreement is that assessment itself must change. “The aim is not merely to make cheating harder, but to make learning more visible and meaningful,” says Adendorff, describing a shift at Stellenbosch “from policing AI use retrospectively to thinking about how we design learning opportunities”.

UP draws a sharper line between low-stakes coursework meant for practice and feedback – formative assessment, in the jargon – and the secure, invigilated exams that actually certify a degree, which must “verify that the student has achieved the required outcomes”.

The Wits policy calls for “AI-resilient assessment” designed to prioritise originality, critical thinking and student engagement. Wits is also writing AI into what it assesses. “With the increased use of AI in industry and work environments, we do need to include the ability to critically and ethically use AI in our learning outcomes,” says Nicole de Wet-Billings, its senior director of academic affairs.

Everyone’s business

Third, all four treat AI literacy as everyone’s business, not a niche skill for computer scientists:

NWU has run a free course completed by more than 3 000 students;
UP embeds AI literacy in a first-year module, informed by its own 2025 survey, which found that AI use is already mainstream among first-years but that responsible use lags behind awareness of the risks;
SU separates “technical” from “professional” AI literacy and wants the latter in every degree; and
The first of Wits’s six principles is to “foster AI literacy” for all staff and students, explicitly “acknowledging that AI access is not equitable in our context”.

There is also a shared insistence that transparency, not surveillance, is the way forward: UP expects students to declare AI use, and Wits attaches a detailed disclosure template to every submission, on the principle that “AI cannot be an author on any work”.

None of the four sees AI as a cheap shortcut, either. Asked whether it lets them teach more students at lower cost, SU and UP answered, in effect, that it’s both: a way to support students at scale, and a significant new cost in licensing, staff development and assessment redesign. At Wits, De Wet-Billings points to the cost of licences to pilot and adopt AI programmes – “new costs that we have not had to provision for in the past” – while NWU is the most optimistic, expecting AI tutor bots to cut costs and improve student success.

The universities diverge most on where and how AI should be governed:

NWU has moved fastest and most formally, becoming what it says is the first university in South Africa and on the continent to approve an official AI policy, which deliberately locates AI governance inside the IT department and is run by a dedicated AI Hub.
“Notably, many of the concerns raised were philosophical rather than technical,” Anné Verhoef, the philosopher who directs it, said in a statement shared with TechCentral.
Wits has taken almost the opposite route, placing strategic leadership with its senate and subcommittees, while its faculties “are leading in designing the parameters of use based on specific tasks”, says De Wet-Billings. It describes the document, which passed through the university’s senate in November 2025 and was approved by its council in June 2026, as a “living” one for a research-intensive institution.
UP has anchored its response in senate-approved guidelines for lecturers rather than a single overarching policy, while SU is still consolidating “several necessary responses in different environments” into a more coherent whole.

They differ, too, on how far to lean on the tools they are integrating. SU has switched Turnitin’s AI detector off entirely; Wits, UP and NWU lean instead on disclosure and redesign, with NWU adding a teaching-first twist: students suspected of misconduct are not immediately referred for disciplinary action but logged on an internal integrity system, “not on their permanent academic record”, given a warning and required to complete a remedial course on responsible AI use.

Read: AI is breaking the link between university degrees and employment

Even AI literacy is delivered differently: NWU keeps its flagship student course voluntary, “as the aim is to promote meaningful engagement rather than compliance”; UP builds it into a required first-year module; and Wits folds it into a broader first-year digital-literacy course, so that, in De Wet-Billings’s words, “AI literacy is not taught in isolation from other literacies”.

Wits sees the danger the other way round. Where the others call their biggest risk the hollowing out of a degree, De Wet-Billings names the opposite: “A major risk for us would be to not engage with AI.”

What is a university, anyway?

What further unites the responses is a willingness to let AI reopen the oldest question in higher education. “AI does not remove the need for universities. It sharpens the question of what universities are for,” says Swart, insisting a degree “must still certify what the student can understand and do, not merely what a tool can produce on their behalf”.

Adendorff puts it almost identically: the university “cannot be reduced to producing assessment outputs or credentials” but exists “to form people who can think, judge, inquire, create, act ethically and contribute responsibly to society”.

Wits states the stakes on the cover of its policy: “Artificial intelligence will not define us. Integrity will.” Verhoef goes further still, warning that professional bodies may one day open their examinations to self-taught candidates, forcing universities “to reconsider their unique contribution”.

That soul-searching is unfolding against a national AI policy mess. South Africa’s draft national AI policy was withdrawn recently after it emerged that parts of it had been drafted with AI, complete with fabricated academic citations – a reminder that the temptation to let the machine do the thinking reaches well beyond the student body. The job of rebuilding it has since fallen to an expert panel chaired by Wits AI researcher Benjamin Rosman, head of the university’s Machine Intelligence and Neural Discovery (Mind) Institute.

The universities’ wager is that the answer is not better surveillance but better design. As Landman puts it, the goal “should not be to win an unwinnable technological race but to establish a renewed contract of trust: one in which AI is used as a scaffold for thought, not a substitute for it”.

Read: Malatsi moves to rescue South Africa’s botched AI policy

And if the sector gets there, it intends to get there together. Asked what sets Wits apart from its peers on AI, De Wet-Billings offers a disarming answer: “We are not doing anything different from other universities. We are learning from and with each other.”

The bet, shared across all four institutions, is that a machine that can write the essay still cannot do all of the thinking – and that a university’s job, now more than ever, is to make sure its graduates still can. – © 2026 NewsCentral Media