Using AI without losing academic integrity: A practical guide for students

Artificial intelligence (AI) has moved into universities faster than most academic rules could be rewritten.  Students can now ask a generative AI tool to explain a difficult judgment, summarise a journal article, suggest research questions, improve grammar or produce an entire essay in seconds.

That convenience creates an important distinction.  Using AI is not inherently unethical, but using it in ways that misrepresent who produced the work, bypass an assessment requirement, or introduce unverified information can undermine academic integrity.  South African universities are increasingly responding with institution-specific rules, declarations and guidance.  For example, UCT’s AI in Education Framework emphasises academic integrity and AI literacy, while Stellenbosch University requires postgraduate students to declare their use of AI in research.

This means there is no single rule, such as ‘AI is allowed’ or ‘AI is prohibited’, that applies to every student, module, and assessment.  The correct answer depends on your university’s policy, your faculty or module rules, the instructions for the particular assessment, and the way in which you use the tool.

This blog explains the main academic integrity risks associated with generative AI, how common AI use has become, what recent misconduct cases can teach students, and the practical steps students can take to use AI as a learning tool without allowing it to replace their own academic work.

 

How widespread is AI use in higher education?

AI use is no longer a marginal issue.  The 2026 Student Generative AI Survey by the Higher Education Policy Institute (HEPI), based on a sample of 1,054 full-time UK undergraduates, found that 95% had used AI in at least one way, and 94% reported using generative AI to help with assessed work.  Twelve (12%) said they had directly included AI-generated text in assessed work, compared with eight (8%) in 2025 and three (3%) in 2024.

These figures are from the United Kingdom and should not be treated as a measurement of South African students.  They do, however, illustrate how rapidly generative AI has become embedded in university studies.  South African evidence points in the same general direction, although national prevalence data remain limited.  A 2025 South African survey of 102 Master’s students in engineering management examined student attitudes and use of AI in higher education, while universities themselves are reporting growing concern about AI-assisted misconduct.  In March 2026, ITWeb reported that UJ, Unisa, UP, and UCT were responding to incidents involving inappropriate use of generative AI.

The important point, therefore, is not that every use of AI amounts to cheating.  It does not.  Rather, AI has become so common that students need to understand the boundary between assistance and substitution before using it in assessed work.

 

What are the main ethical problems with AI in academia?

The central concern is not the technology itself.  The problem arises when AI changes the authorship, accuracy, or integrity of work submitted for academic credit.

Submitting AI-generated work as your own

The clearest risk arises when a student asks an AI tool to produce an essay, assignment, discussion response, research section, or examination answer and submits that output as if it were their own work.  Even when the wording is technically new and not copied from an existing publication, the student may still be misrepresenting the authorship of the work.

This is why students should not assume that avoiding conventional plagiarism means AI-generated work is acceptable.  The Council on Higher Education has stressed the need to rethink academic writing and integrity in the era of generative AI rather than treating the issue only as traditional plagiarism.

Fabricated references and inaccurate information

Generative AI systems can produce confident answers that are inaccurate.  They may invent case names, journal articles, quotations, page numbers, legislation or authors that appear entirely plausible.  This is commonly described as an AI ‘hallucination.’

The problem is already visible in South African academia.  In May 2026, Stellenbosch University warned staff and students about fabricated and hallucinated references, noting that the university had encountered several cases involving fabricated references in course and research assignments, theses and journal publications since 2024.

For a law student, the risk is particularly serious.  A fictitious judgment or incorrect statutory provision can change the legal conclusion entirely.  AI should therefore never be treated as the authority for a legal proposition simply because its answer sounds convincing.

Using AI where an assessment prohibits it

An AI tool may be acceptable in one module and prohibited in another.  For example, a lecturer may allow the use of AI for brainstorming but prohibit the use of generated text in the final submission.  Another assessment may require students to use AI and critically evaluate its output. In a closed-book examination, using an AI assistant may amount to the use of an unauthorised aid.

The University of Pretoria’s teaching guidance shows why assessment-specific instructions are important, recommending that lecturers state whether generative AI is prohibited, permitted with limitations and appropriate acknowledgement, or required as part of the learning outcome.

Failing to disclose AI assistance

Even when the use of AI is permitted, a student may still be required to disclose how it was used.  A declaration can distinguish legitimate assistance from hidden substitution of the student’s own work.  Stellenbosch University’s current approach, for example, requires postgraduate students to declare their use of AI and encourages students and supervisors to agree on responsible use during the research process.  Its interim assessment guidance also contemplates declarations identifying the tool used, where it was used and how it contributed to the work.

Privacy, confidentiality and sensitive research material

Students should also think carefully before uploading information into a public AI system.  Interview transcripts, unpublished research, participant information, confidential workplace documents and personal data may be subject to research ethics requirements, confidentiality undertakings or privacy rules.

If you would not upload a document to a public website, do not assume that it is appropriate to paste the same document into an AI chatbot.  This is why it is important that you first check the tool’s data practices and your institution’s rules before using sensitive material.

Over-reliance and the loss of learning

There is also a less visible ethical problem: AI can produce a polished answer before the student has developed the skill the assessment is designed to test.  If a student repeatedly outsources reading, reasoning, drafting, or problem-solving, the submitted work may improve while the student’s own competence does not.

A useful test is simple: after using AI, can you independently explain and defend every argument, source, and conclusion in your submission?  If the answer is no, the tool may have replaced too much of the learning process.

 

What can recent academic misconduct cases teach students?

Recent cases show two equally important lessons.  Universities can impose serious consequences for unauthorised AI use, but an AI detector should not automatically be treated as proof of misconduct.

Case study 1: ChatGPT used during an examination

In a 2026 University of Toronto disciplinary case, an invigilator discovered a student using a smartphone with ChatGPT open after images of test questions had been uploaded to the tool.  The disciplinary panel found the student guilty of using an unauthorised aid.  Because the student had a prior academic offence and the conduct was considered planned and serious, the sanctions included a zero for the course, a five-year suspension and a recommendation for expulsion.

The lesson is straightforward, clearly showing that using AI during an assessment that prohibits outside assistance is not simply an innovative use of technology but can be treated like other forms of examination cheating.

Case study 2: Undeclared AI use in a dissertation

A 2025 case summary published by the UK’s Office of the Independent Adjudicator for Higher Education concerned a student who acknowledged during a viva that AI had been used to help write a dissertation but had not declared that use on the required form.  The institution found academic misconduct and awarded a mark of zero.  The subsequent complaint was ultimately found justified because of a procedural irregularity in the institution’s process, not because undisclosed AI use was necessarily acceptable.

The practical lesson is that disclosure requirements matter.  If an institution asks you to declare your use of AI, honesty after submission may not undo the failure to comply with the original requirement.

Case study 3: Why AI detection alone can be problematic

Another 2025 OIA case involved a postgraduate student whose essay was flagged for suspected AI use and contained apparently fabricated references.  The provider initially failed the student for the module, but the complaint was later settled after concerns arose about whether the institution had properly engaged with the student’s evidence and explanations.

South African universities are also reassessing automated detection.  UCT discontinued access to Turnitin’s AI Score from October 2025 because of concerns about false positives and false negatives, and the University of the Free State discontinued AI detection software across faculties from July 2026 amid similar concerns about reliability, bias and due process.

Students should therefore take two messages from these developments.

1. Do not assume that AI misuse cannot be detected; and
2. Remember that an AI-detection percentage is not the same thing as a conclusive finding of misconduct because fair academic processes should consider the surrounding evidence.

 

How can students use AI ethically?

The safest approach is not to ask, ‘Can I use ChatGPT?’ but rather, ‘What am I allowed to use AI for in this specific assessment, and how do I remain the author of the work?’

Practical do’s

• Check the rules before you start:  Read your university policy, faculty guidance, module outline and assessment instructions. If they differ or are unclear, ask the lecturer or supervisor before using AI.
• Use AI to support learning:  Appropriate uses may include asking for a simpler explanation of a difficult concept, generating practice questions, brainstorming possible research directions, or receiving feedback on clarity, where your institution permits this.
• Verify every factual claim:  Open and read the actual judgment, statute, journal article, or book before relying on it.  Never cite a source merely because an AI tool supplied the citation.
• Keep a record of your use:  Where appropriate, save prompts, outputs, and notes that show how AI contributed to your work.  This can help you complete an AI declaration accurately and demonstrate your own research process.
• Acknowledge AI when required:  Follow the citation or declaration method prescribed by your university, faculty, lecturer, or supervisor.
• Protect confidential information:  Remove identifying or sensitive information and do not upload protected research data unless the tool and intended use have been approved.
• Remain able to defend your work:  You should understand every argument and be able to explain how you reached your conclusions without relying on the chatbot.

Practical don’ts

• Do not submit generated text as your own when this is prohibited:  Rewording AI output does not necessarily eliminate the underlying academic integrity problem if the intellectual work was outsourced.
• Do not invent or trust references blindly:  If you cannot locate the source independently, do not cite it.
• Do not use AI in a closed assessment unless expressly permitted:  Using an AI assistant during an examination can amount to unauthorised assistance.
• Do not upload another person’s work or confidential data casually:  This includes classmates’ assignments, unpublished manuscripts, client documents, and identifiable research participant information.
• Do not assume one rule applies everywhere:  Permission in one module does not automatically extend to another module, lecturer, dissertation, or examination.
• Do not use AI to avoid developing the skill being assessed:  If the purpose of the task is to test legal reasoning, writing or research, the final intellectual work must remain yours.

A simple test before submitting AI-assisted work

Before submitting an assignment, dissertation chapter, or research output, ask yourself:

• Was AI permitted for this task?
• Have I complied with any disclosure or citation requirements?
• Can I identify and verify every source independently?
• Have I checked the output for factual, legal, and contextual errors?
• Is the argument genuinely mine, rather than an AI answer that I have merely edited?
• Could I explain and defend the work in a viva, a tutorial, or a meeting with my supervisor?
• Have I protected confidential, personal and research data?
• If you cannot confidently answer yes to these questions, review the work and the applicable academic rules before submitting it.

 

Resources for students

AI policies are changing quickly, so students should prioritise current institutional guidance over informal advice on social media.  Useful starting points include:

• Council on Higher Education: plagiarism and generative AI – South African higher-education guidance on academic writing, plagiarism and generative AI.
• Stellenbosch University AI resources – declarations, guidance on responsible AI use and links to further training.
• Wits University AI Policy – institutional principles for responsible, ethical and productive AI use.
• Wits ReadWrite Studio – academic writing support, consultations and guidance on responsible generative AI use.
• Unisa Academic Integrity resources – academic integrity training addressing generative AI and other forms of academic misconduct.

Students at other institutions should also check their own university library, writing centre, postgraduate office, research ethics office and faculty or module guidance.  Where the rule remains unclear, the safest resource is often the person responsible for the assessment: ask before you submit.

 

Conclusion

Academic integrity in the age of AI is not about pretending these tools do not exist, but rather about ensuring that technology strengthens learning without obscuring authorship, weakening scholarship, or replacing the skills a qualification is meant to represent.

The most useful question for students is therefore not how much AI they can get away with using, but how they can use it while remaining accountable for the work submitted under their name.  Policies will continue to evolve, but that principle is unlikely to change.

 

About the author

Theshaya Naidoo is a PhD (Law) Candidate and Canon Collins Scholar.  Her research focuses on 'The Legal & Ethical Implications of Neurotechnology on the South African Criminal Justice System'.  She holds an LLM in Medical Law, where her thesis focused on 'The necessity of sui generis AI regulation in South Africa'.  Theshaya attended Gawie le Roux Institute of Law’s Attorneys’ admission exam courses and law school.

 

Last updated on 16 September 2026.