Artificial intelligence has reached a point where it can credibly complete almost any written assignment in an undergraduate curriculum, according to a new report from an ad hoc committee at the Massachusetts Institute of Technology (MIT). The finding, released this week, covers essays, mathematics and science problems, proofs, and coding assignments, indicating that the capabilities of large language models and related AI systems now span nearly the entire range of academic exercises given to college students.
The report was prepared by MIT's ad hoc AI committee, a group convened to assess the impact of generative AI on the institute's educational mission. While the committee did not explicitly frame the findings as an indictment of current assessment methods, the implications are profound: if AI can produce credible solutions to almost any assignment, then the traditional take-home exam and problem set may no longer serve as reliable measures of student knowledge or skill.
Key Findings of the MIT Report
The report's central claim is broad and unambiguous. AI models are now able to generate responses that would receive passing or even high grades across a wide variety of undergraduate coursework. This includes not only standard essays and short-answer questions but also complex mathematical derivations, scientific problem sets, formal proofs, and computer programming tasks. The committee found that the technology produces credible solutions across all of these categories, a notable acceleration from just a few years ago when AI struggled with anything beyond basic text generation.
The significance of this finding goes beyond the obvious concerns about academic dishonesty. The report suggests that the very nature of what it means to complete an assignment is being called into question. If an AI can do the work, then the assignment may no longer be testing the skills that educators intend to measure. This is not a hypothetical issue; it is a current reality that MIT and other universities must confront as they design curricula and assessment methods for the coming decades.
Campus Culture in Transition
Perhaps more interesting than the technical capabilities of AI is the report's observation that the technology has already driven what it describes as 'major shifts' in campus culture. These changes have occurred in under three years, a remarkably short period for any technological development to reshape the social fabric of a research university.
Attendance at office hours is down, according to the report. Participation in online discussion forums has also fallen. The committee cites anecdotal evidence of fewer study groups forming in dormitories and libraries, a traditional hallmark of collaborative learning at institutions like MIT. None of these changes are necessarily the result of cheating; rather, they suggest that students are turning to AI tools as a substitute for human interaction, whether for help with difficult concepts or for feedback on their work.
This cultural shift is not unique to MIT. Universities across the United States and around the world have reported similar trends since the public release of powerful AI chatbots. The report's contribution is to document these changes systematically and to connect them directly to the spread of AI tools in an academic setting.
Institutional Responses
Other universities have already begun to react to the challenges posed by AI. The University of Chicago's law school, for example, banned phones and laptops in first-year classes, a move aimed at reducing distractions and limiting the potential for AI-assisted participation. Princeton University has taken a different step by dropping an honour code that had been in place for more than a century, acknowledging that the assumptions about student integrity underlying such codes are no longer tenable in an era of ubiquitous AI.
These responses reflect a broader divide in how institutions are approaching the problem. Some are doubling down on supervision, requiring students to complete work in controlled environments where AI use can be monitored. Others are reconsidering the very nature of assignment design, focusing on skills that AI cannot easily replicate, such as oral presentation, hands-on laboratory work, and real-time problem solving.
The direction of travel, according to many observers, is toward greater supervision. If you cannot trust the work, the logic goes, then you watch the person doing it. This is a very old answer to a new question, echoing the invigilated exams that have been a staple of education for centuries. But the move toward supervision raises its own concerns about privacy, equity, and the student experience.
The EU AI Act and Educational Oversight
That is where European law starts rather than ends. On this continent, the watching itself is the regulated activity, not the cheating. The European Union's Artificial Intelligence Act, which entered into force in 2024, takes a highly structured approach to the use of AI in education. It singles out education as one of the areas that receives the strictest treatment short of an outright prohibition.
Under Annex III of the AI Act, systems used for monitoring and detecting prohibited behaviour of students during tests are classified as high-risk. This classification applies alongside admissions systems and the evaluation of learning outcomes, meaning that universities and ed-tech companies must meet stringent requirements for transparency, accuracy, and human oversight before deploying such tools.
The obligations associated with these high-risk classifications were due to take effect this month. However, the European Commission's Digital Omnibus package, a broader regulatory adjustment affecting multiple pieces of digital legislation, pushed the application date back to 2 December 2027. This delay buys institutions and their software suppliers another sixteen months to adapt their systems and processes to the new legal framework.
Notably, the prohibitions contained in the AI Act were not pushed back. Emotion recognition in education has been banned outright since February 2025, meaning that any AI system that attempts to infer a student's emotional state during learning or testing is now illegal in the EU. The European Commission can inspect models and fine providers for violations, giving the prohibition real enforcement teeth.
The Broader Implications for Higher Education
The contrast between the American and European approaches is instructive. In the United States, the debate is largely framed around institutional autonomy and academic integrity. Should individual universities be allowed to monitor students' devices and activities during exams? Should they be permitted to use AI to detect AI-generated work? These questions are being answered at the campus level, with widely varying results.
In Europe, the debate has been settled at the regulatory level. The AI Act establishes a common set of rules that apply across all member states, overriding national and institutional discretion in key areas. The ban on emotion recognition is an absolute limit, while other monitoring tools are allowed only if they meet rigorous standards of proportionality and human oversight.
For universities, the challenge is to navigate this complex regulatory landscape while maintaining academic standards. Some are experimenting with AI-resistant assessment methods, such as oral exams and project-based portfolios. Others are integrating AI literacy into the curriculum, teaching students how to use the tools responsibly and transparently.
The MIT report adds an empirical dimension to these debates. It shows that AI's impact on education is not speculative or future-oriented; it is happening now, and it is affecting not just cheating but the entire culture of learning. As other institutions look to MIT's findings, they will have to decide how to balance the benefits of AI, such as personalized tutoring and instant feedback, against the risks of eroding human connection and undermining trust.
The European Union's approach offers one possible model. By distinguishing between acceptable uses of AI, such as adaptive learning systems, and unacceptable uses, such as emotion recognition, the AI Act creates a clear framework for innovation within boundaries. The delay of the high-risk obligations may provide temporary relief, but the direction is unmistakable: in the EU, the watching of students is a regulated act, and the burden of proof lies with the institutions and companies that wish to engage in it.
For now, universities on both sides of the Atlantic are left to grapple with a fundamental question: how do you assess learning in an age when machines can produce credible answers to almost anything? The MIT report does not provide an answer, but it does force the question to the top of the agenda. And as the technology continues to improve, that question will only become more urgent.