(369) Times Higher Ed and An Important Paper
Plus, ICAI calls for 2026 Conference Proposals.
Issue 369
Subscribe below to join 4,717 (-2) other smart people who get “The Cheat Sheet.” New Issues every Tuesday and Thursday.
The Cheat Sheet is free. Although, patronage through paid subscriptions is what makes this newsletter possible. Individual subscriptions start at $8 a month ($80 annual), and institutional or corporate subscriptions are $250 a year. You can also support The Cheat Sheet by giving through Patreon.
Times Higher Ed on New Paper on AI and Integrity
Times Higher Ed (THE), which continues to lead all outlets in covering academic integrity, has a story on a new academic paper from a trio of authors in Australia — Thomas Corbin and Phillip Dawson from Deakin University, and Danny Liu from the University of Sydney.
I’m going to touch on the coverage first, the paper itself second.
To start, and probably most importantly, I love that THE continues to write about integrity. It’s a topic most publications ignore, or cover quite badly, even with great damage. Further, it’s quite rare that any outlet gives space to academic research on integrity, which this article does. For that, great work.
Even so, the article itself is kind of a mish-mosh, and I fear it will be taken incorrectly. The headline is:
Universities’ AI safeguards promote ‘enforcement illusion’
Immediately, you may realize this can go one of two ways.
One way, the enforcement illusion is because schools do not want to enforce AI guidelines. They issue rules but do not enforce them, or even inspect the results, because they do not want to. The other way is that schools do not enforce AI guidelines because it is impossible to do so.
It’s on this important point that the authors of the paper and I part company, dramatically and soon. But sticking with the THE story, the opening paragraph is:
Universities’ efforts to “AI-proof” assessments are built on “foundations of sand” because they rely on instructions that are “essentially unenforceable”, a new paper argues.
Fair, I think. There’s no objective line between using AI to “polish” or “correct grammar” on an assignment and using AI to rewrite parts of it. People’s views of those lines are different. Therefore, telling students editing is fine, but using AI text is forbidden, is pointless. This point is made in the research paper.
And I really agree with this overview, from the paper authors, in the THE coverage:
The researchers liken universities’ use of the traffic light system – a three-tiered classification where red means no AI use, amber indicates limited use and green allows unrestricted use – to “real traffic lights” without detection cameras or highway patrols.
“Traffic lights don’t work just because they are visible,” the researchers explain in the journal Assessment & Evaluation in Higher Education. “Educators might assume these frameworks carry the same force as actual traffic lights, when in fact they lack any meaningful enforcement mechanism. This creates a dangerous illusion of control and safety.”
One hundred percent correct. Rules, without any supervision or enforcement, are just junk words. As the paper authors also include, junk word rules not only do not work, they usually create harm.
Also from the coverage:
Lead author Thomas Corbin, a fellow at Deakin University’s Centre for Research Assessment and Digital Learning, said the first step for educators was to acknowledge that instructing students about AI use – and imagining they will comply – was “just not going to work”.
On this, I agree thoroughly as well. Assuming voluntary compliance with rules related to using AI is foolish — my word, not theirs. When there are rewards for using AI, and no immediate negative consequences, expecting compliance is fantasy thinking.
Finally, from the article:
Educators should use “structural” approaches that “build validity into the assessment design itself rather than trying to impose it through unenforceable rules”, the paper argues. Examples could include requiring students to produce essays under supervision, giving students random questions during “interactive” oral assessments, or requiring tutor sign-off on lab work.
I remain skeptical that assessment redesign will significantly limit unauthorized use of AI or other cheating tools. As has been shown and written about already, there is no assessment that is AI-proof (see Issue 352).
Although, at the same time, these specific suggestions would unquestionably deter some types of AI-related misconduct. The more places for checks and supervision, the harder it becomes to succeed at cheating.
Anyway, good on THE for keeping this issue and these conversations on their pages.
The Paper on “Enforcement Illusion”
The paper covered by THE and discussed above is here.
Trying to avoid hyperbole, it’s the best and also the worst paper I’ve seen on this issue — perhaps ever.
This is the first line from the abstract:
Generative AI (GenAI) challenges assessment validity by enabling students to complete tasks without demonstrating genuine capability.
Seems obvious. And you will find no louder voice of agreement than mine. This is the problem, singular.
The Important and Essential
The paper does a compelling job of reviewing the various frameworks that many colleges and universities are using to address AI use. Nearly all of those, the paper argues, are some form of “traffic light” type of guidance. From the paper:
We find that educational frameworks frequently borrow the language of socially familiar structural systems (like vehicular traffic lights) while lacking their actual enforcement capabilities, creating an illusion of assessment security.
Fair. Also:
We propose that many of the current assessment frameworks depend primarily on what we term discursive changes to assessment, modifications that rely solely on instructing students about permissible AI use, which students remain essentially free to follow or ignore. We contrast these with a different category, which we call structural changes to assessment. These are modifications that reshape the underlying mechanics of the assessment tasks themselves, thereby directly influencing or constraining how students can interact with GenAI.
No problem whatsoever. Agreed.
The authors say that the flaw in such discursive settings is the nearly complete reliance on voluntary compliance. For example, citing a supporting work, the authors say:
there is little reason to think that students will adhere to these instructions and use AI in ways that correspond with the level selected by the educator.
Without supervision and enforcement, there’s little reason to expect any compliance whatsoever. Consider:
There is, however, an implicit assumption that these levels of appropriateness will be adhered to by students. In fact, within the assessment of learning scope of this paper, the success of these approaches is entirely dependent on this adherence. Communication and ‘transparent conversations’ (albeit important in many ways), accompanied by indications of how AI ‘may’ or ‘should’ or ‘must’ be used, potentially provides both a false sense of security and present a risk to assessment validity. We therefore suggest that any assumption regarding student compliance is problematic.
Success is “entirely dependent” on adherence, they say. “Any assumption regarding student compliance is problematic,” they say.
I say, no kidding.
Here, the hammer strikes the head of the nail:
[these policies] attempt to elicit compliance through language alone, without corresponding mechanisms to enforce those boundaries.
Yup. Just imagine, for a moment, a setting in which drunk driving was not illegal. No law enforcement checked for driving while under the influence, no one ever got a ticket for it, nothing. Instead, we used only a system of voluntary compliance with a green-yellow-red code to advise people when they should or should not drive after drinking.
The idea is absurd. And the authors are right to say so. It’s important.
The authors add:
such approaches rely heavily on student compliance with instructions, a reliance which opens both assessments to vulnerabilities as well as educational institutions to reputational risk
Slow. Clap.
The authors also argue that assessments given with mere direction related to using AI are likely to be invalid, or even harmful:
from a validity standpoint any change which is merely discursive and not structural is likely to cause more harm than good.
I still agree. Absolutely.
The paper also cites institutions at which the policy is not a “traffic light” approach but one of required disclosure of AI use. They wrote:
many higher education institutions rely on students declaring their use themselves
The problem with this approach is the same, however. Students simply don’t do it.
Continuing:
Other institutions, such as King’s College London, have introduced a mandatory AI use declaration on assessment cover sheets, where students indicate whether they used generative AI and briefly describe its application (King’s College London n.d.). However, compliance has been an issue. One study at King’s found that up to 74% of students did not complete the AI declaration appropriately, with many fearing that admitting AI use might be perceived negatively
That’s new data, at least to me. And it’s unsurprising.
Why would anyone willingly flag anything that someone could later determine to be a problem? Imagine the IRS asking tax filers to tell them anything that could be a problem with their returns. It’s bonkers.
Every bit of that is important. Educators and school leaders need to hear that if their AI policy is discursive, directive, and lacks enforcement — or one that relies on self-reporting — there is essentially no policy. Or worse.
The authors continue to say that the path forward is to redesign assessments — “structural assessment redesign.” I already mentioned that I find this to be wishful thinking, so, I won’t bang on that again.
The Wrong and Discrediting
After all that good and important stuff. If the issue is lack of enforcement, the answer seems easy — enforce it. If a school or instructor prohibits AI use in some way, do something about it.
Instead, the authors surrender their credibility by insisting that enforcement cannot happen because AI cannot be detected. That, of course, is simply untrue. Proven untrue. Wrong.
They say, for example, that bans on AI use:
would only be viable if reliable AI detection technology existed, which it currently does not
False.
current detection tools are fraught with false positives and negatives, creating uncertainty and mistrust rather than clarity and accountability. In practice, educators often have no feasible way to confirm whether a student complied with instructions regarding AI use, making compliance essentially unenforceable.
False.
What’s maddening about these false statements is that I know the authors know better. At least two of them subscribe to, and read, this newsletter. They know that there is not a single study anywhere that shows AI detectors — every single one of them — are inaccurate and unreliable. Not one.
The good detection systems — Turnitin, Pangram, Watson, and Rosalyn to name four — have routinely shown accuracy rates well above 90%. Most, with zero false positives at all. To say that educators “have no feasible way” to detect AI is disingenuous.
The paper outright says that past enforcement structures related to student work had:
traditionally been supported by shared values around academic integrity and the implicit threat of detection through plagiarism checkers or stylistic inconsistencies. GenAI fundamentally disrupts this arrangement by making detection unreliable while maintaining the appearance of original work.
This literally says that AI, by its very nature, cannot be reliably detected. It is not a statement based in reality.
The paper also, at one point, says:
perfect detection remains technologically unfeasible
Perfect. The standard is perfection — a standard no other detection system in the world attains. Go ahead, name any detection system in the universe, technology or human, that is perfect. Go ahead. I’ll wait.
I may be beating this analogy, but on my last flight, the airport bag screening thing flagged my bag. A TSA agent called me over, asked me a few questions, opened the bag and dug around. Whatever he was alerted to was not a problem, obviously. But imperfect. False positive. Unfeasible. Can’t use them anymore.
And, by the way, no one accused me of anything.
Back to the paper. On about page five of 11, there’s this:
Relying on these unreliable tools harms students through false accusations, while also damaging institutions and teachers by promising them a level of security these systems simply cannot deliver.
And at that point I did something I never do. I stopped reading.
That sentence is so wrong, demonstrates such willful misunderstanding, that it cannot be taken seriously.
The tools harm students through false accusations. I just can’t.
For the millionth time, detection systems — AI or any other — do not make accusations. Not ever.
No metal detector at any airport in the entire history of aviation has ever accused anyone of terrorism. No smoke detector in the entire history of fire has ever accused anyone of arson.
This is not complicated. And the only plausible reason for anyone to say such a thing, in my opinion, is deliberate misinformation. Or perhaps unchecked ignorance. But, again, I have every reason to believe that the authors of this paper are not ignorant.
I understand that the authors want people to redesign their assessments. That’s fine. There are many good reasons to do so. But they don’t have to misrepresent — falsely represent — a perfectly viable alternative in order to make their case.
If lack of enforcement in AI policy is a problem, and it is, enforce AI policy. It can be done. Saying otherwise is untrue, and corrosive to the core of integrity as an academic imperative. And literally.
2026 ICAI Conference in Denver Calls for Proposals
The International Center for Academic Integrity will hold its 2026 Conference in Denver and has opened its call for proposals.
It’s the essential gathering of the year. I hope you’ll make plans to attend and prepare to share.
Reposting this Request
I’m working on a new, major project with direct implications for the value of authentic work, not just in academia but much more broadly. It’s a challenge I find acute and long-lasting.
Since you read The Cheat Sheet, you may agree that human work — especially writing — has deep value. For creator and consumer alike.
If you agree, or even if you’re one of the three people who just likes me personally, I need a little help. I need to connect with people who can help get this project moving. Specifically, I am seeking connections and introductions to:
Angel investors or supporters who can invest $10,000 or more in this important, and likely profitable, enterprise. We need to raise $150,000 rather quickly.
Publishers, agents, or media executives with experience in their fields and strong networks who care about the value of human writing. Founders of media companies, book agents or publishers, executives at academic journal companies or TV studios — people who buy, publish or otherwise share written work. I’d like to speak with them about being Advisors.
High profile authors who also care deeply about the authenticity and human value in writing.
If you personally are in any of those categories, please say hello. If you know anyone in any of those categories and can make an introduction, I am asking.
If you’d like to know more about what I’m working on, please ask. Although it’s not ready — hence the investment-seeking — I am happy to share.
Thank you.