There is no escaping the prevalence with which A.I. is increasingly pervading our lives.  Young people growing up now may be unable to comprehend a world in which A.I. was not with us, the silent partner in their individual life experiences.  COVID-19 brought many changes to education, never mind society in general.  What was previously viewed as something that could only be done in person, is now frequently done remotely.  Although most universities have now long returned to in-person lectures, seminars and lab sessions, examinations are still predominantly undertaken remotely.

There have always been 2 pillars of concern in any assessment exercise – (a) Does the student genuinely know the content (as opposed to seeking the answer from sources other than their own minds)? (b) Does the student understand the issues / topics in question (or are they just parroting back what they learnt by rote)? & (c) Were any parts of their submitted work written by someone else?  The latter issue is why students must reference any sources relied upon for their answer, and in particular to properly attribute quotes.  But what if the source of the ‘knowledge’ is not a person?

Universities have sought to adapt to this trend with varying success.  In many universities there is now an acceptance that many students will use A.I. to some degree in the completion of their written work, leading to the adoption of student ‘declarations’ for students to specify in what way and to what extent they used A.I. in the completion of their work.  At first blush this seems like a sensible approach, and one standing in stark contrast to the near uniform refusal to return to in-person examinations, which one would have expected to be almost the first step to resisting the risk of A.I. influencing assessment results.

The difficulty is two-fold: (a) recent surveys suggest that since the last academic year there has been a swing in students declaring A.I. use from over 90% down to 32% in some institutions; and (b) how universities have tried to identify A.I. use.  We have all seen the reports in the media of innovative strategies employed by lecturers in an attempt weed out undeclared A.I. use; but with the increased acceptance of A.I., the question still remains as to whether students have accurately declared the level of A.I. use in their work.  There are now a number of A.I. tools that can be used in conjunction with one another, for example one to produce a base essay answer and another with the sole function of ‘humanising’ it.  What then are universities to do?

We looked previously at how the MIT Sloan School has noted recent reports show that A.I. detectors have “…high error rates and can lead instructors to falsely accuse students of misconduct”.  I have encountered a number of instances where examiners with prior knowledge of a student have accused them of academic misconduct because of the misconceived view that the submissions ‘too good’ to have been produced by that student.  There is little consideration given to the possibility that a student could have put in extra effort or had addressed issues that were impeding their academic performance previously.

Equally, many universities are using Turnitin to try to detect A.I., but that is not what it is designed for.  It is a plagiarism detection tool wildly adopted across higher education providers in the UK.  Yet it is being used as the initial ‘detection tool’ from which the A.I. witch hunts begin, again with students being accused of attainment beyond lecturers’ expectations.

Although some institutions publicly report a very low level of A.I.-related academic misconduct (for example, one reputable London-based university reported only 32 students being investigated for A.I. academic misconduct in 2024/25, resulting in 17 students being disciplined and 5 having their studies terminated); the unfortunate fact is that many students will be facing this gauntlet as results now begin to release for their winter assessments.

Universities must understand the limitations of any tool they use to help identify A.I usage, and it is vital that they give due regard to students’ explanations for why their work may have improved.

One of my recent clients unfortunately found themselves in just such a scenario.  They had indeed achieved a massive increase in their academic performance from prior years.  However, they were able to show that the improvement in their results was due to a combination of finally receiving treatment for previously undiagnosed ADHD and a huge amount of additional work being put in by them.  They were able to evidence this, produce a variety of prior drafts, as well as a clear evidence trail showing when and how they had used A.I. (which they had fully declared in the correct form – in this case, essentially to help plan the essay and to grammar check).  Their work was initially flagged by Turnitin for plagiarism, which led the university to initiate an investigation for academic misconduct based upon undeclared A.I. use.  Despite the student’s evidence, they were found guilty of academic misconduct and told they had to repeat the year, and their stage 1 academic appeal was unsuccessful.  It was only with legal assistance and a stage 2 academic appeal that the university finally accepted the student’s position and allowed them to progress in their course.

For students it is imperative that any A.I. is correctly declared (where its use is permitted by the university), and that a clear evidence chain is kept to show that the work is indeed theirs.  Due to the potential for significant implications of an accusation of academic misconduct, it is one instance where receiving appropriate advice can often change the game.  One needs to understand the university’s regulatory framework, how best to present the evidence available, and what gaps there may be in it.

This is not an issue with an easy answer for universities or students, nor is it one that is going to disappear any time soon.  The best advice that can be given to either side in this equation is to never forget the black and white of the regulations, and the requirements of procedural fairness and natural justice.  Only in this way will it be possible to find a correct balance to ensure that whilst the wolf may be firmly wedged into the corner of every lecture hall, they remain chained, visible, and (hopefully) eventually domesticated.

There is no easy solution to AI in assessment. This issue will not disappear.

What can be controlled is adherence to:

  • Regulatory clarity
  • Procedural fairness
  • Natural justice

Key Takeaways for Universities and Students

For universities:

  • Understand the limitations of detection tools;
  • Avoid assumptions based on prior performance;
  • Engage meaningfully with student explanations; and
  • Apply regulations consistently and fairly.

For students:

  • Declare AI use accurately (where permitted);
  • Keep clear evidence trails;
  • Understand the applicable regulations and policies; and
  • Seek advice early as the consequences of misconduct findings are severe, and challenging a negative decision can a long uphill struggle.