// NATURE NEWS — SPAZIO & SCIENZA
From Trojan horses to AI-proof exams: how professors are tackling students’ AI use
Katarina Zimmer is a freelance journalist in Berlin.
Search author on:
PubMed
Google Scholar
Illustration: Claire Welsh/Nature/Adapted from Getty
The use of large language models (LLMs) is soaring among students worldwide thanks to the tools’ growing ability to solve problems and generate text and code. In a 2026 survey of 1,054 UK undergraduate students by the Higher Education Policy Institute in Oxford, UK, roughly 94% said that they use generative artificial-intelligence tools to help them with assessed work, and 12% directly inserted AI-generated text into their coursework1. In a study published in May, from survey data of more than 95,000 students at 20 US universities, the authors estimated that 9% of students used AI on their coursework, despite knowing that this broke the rules2.
Educators worldwide are seeing suspicious signs of AI use — from hallucinated references to punctuation styles that are typical of AI tools — in take-home coursework and online examinations, fuelling worries that the technology is giving some students an unfair advantage over those who don’t use it. At one point, people were submitting essays in which “50% of the reference literature did not exist in real life”, says Nikita Bezrukov, who teaches linguistics and communication at the Massachusetts Institute of Technology in Cambridge. At the same time, many professors are increasingly recognizing the benefits and inevitability of AI use in students’ education and future careers, and see a need to teach young people to use the technology wisely.
These realizations are forcing many educators to rethink conventional assessment approaches, both to accommodate AI as a tool and detect its misuse. Nature’s careers team spoke to educators around the globe about some of the approaches they and their institutions are taking to revamp assessment.
That said, “I don’t think anybody has a ‘right’ answer right now”, says computer scientist Nicholas Mattei at Tulane University in New Orleans, Louisiana. “Things are changing pretty rapidly.”
“For an assignment in my history-of-technology class, students have to identify sources, use three LLMs to summarize them and critique what the models say. Then, they make infographics and write an essay, for which AI use is allowed. They get extra points if they notice when the infographics don’t accurately reflect the sources or if the LLMs hallucinate references. The goal is to prompt more critical reflection on what comes out of LLMs.” — Nicholas Mattei.
“Instead of writing an essay about Adam Smith’s 1776 economics book The Wealth of Nations, which is what I used to ask students to do, I got them to build a virtual representation of his theory using AI agents acting as traders in a marketplace. We then explored various potential situations — such as what happens if people become untrustworthy. Students analysed the transcripts produced by the agents and wrote about it. It was easy to spot whether students asked AI to do all these steps in one go because, for instance, it just made stuff up rather than quoting from the transcripts.” — Daniel Silver, sociologist at the University of Toronto Scarborough, Canada.
“The exam for my web-development course is open book and open internet, and AI is part of the internet. But I’m assessing a skill that AI doesn’t have — how to build good, sustainable web applications. I ask inherently open questions that have multiple answers, such as “This website doesn’t work, why?” and “Why is this website slow?” If you put this into AI chatbots, they just answer questions without sound methodology, and not even always accurately.” — Ruben Verborgh, computer scientist at Ghent University, Belgium.