I'm a graphic designer who specializes in AI, and I've spent the last four-plus years working in edtech. I use these tools every day. I know how fast they produce something that looks finished.
Education runs in my family. My mom teaches high school. My dad teaches at university and in postgraduate programs. During my design degree I worked as a student teaching assistant, so I've seen the classroom from both sides of the desk. The question of how you actually know someone learned something has been around my whole life.
AI makes that question a lot harder to answer.
In a blind study published in PLOS ONE in 2024, researchers at the University of Reading submitted exam answers written entirely by GPT-4 on behalf of 33 fictitious students across five undergraduate psychology modules. 94% of the AI submissions went undetected, and across modules there was an 83.4% chance the AI submissions would outperform a random selection of the same number of real student submissions.
It's happening at scale, too. As EFE reported on July 31, 2026, UNAM, Mexico's largest university, announced an additional in-person exam for selected applicants after detecting statistical anomalies and signs of irregularities, including suspected AI use, in its first fully online admission process. La Nación reports that more than 158,000 online exams came under suspicion.
As a designer, this doesn't surprise me. A polished output and the thinking behind it are two different things. AI is very good at the first one.
The reflex has been to detect. The data says it's backfiring. The 2026 Studiosity and YouGov Global Wellbeing Survey of 10,330 university students found that 77% of students who use AI feel anxious about being wrongly flagged or falsely accused of cheating by detection tools.
MIT got to the same place from a different direction. In its August 13, 2026 report, MIT's Ad Hoc Committee on AI Use in Teaching, Learning, and Research Training recommended against relying on AI detectors. It warned that they can mistake the writing of non-native English speakers and neurodivergent students for AI, and that policing AI use builds an atmosphere of distrust between instructors and students. The report also notes that MIT's own disciplinary committee doesn't consider detector output alone sufficient evidence.
The same MIT report describes decreased attendance at office hours, less participation in online discussions, and, from what the committee heard anecdotally, fewer in-person study groups. Drawing on recent research, it also warns about "cognitive surrender": a correct chatbot answer creates the illusion of learning, and students fall back on AI at the first hint of struggle.
The human support around students is shrinking as well. In the same Studiosity and YouGov survey, only 40% of students globally had access to a mentor in 2026, an 18-point drop from the year before.
The cheating story gets the headlines. This is the part that will cost us more.
Almost everyone is using AI now. The Digital Education Council's AI in Higher Education Global Survey 2026, drawn from 45,398 responses across 35 countries, found that 88% of students use AI in their learning and 77% of faculty use it in their teaching. What hasn't caught up is everything around that. The same survey found that 57% of students say their assessments come with inadequate AI guidance, and only 29% believe their instructors are well equipped to guide them on AI use. In the US and Canada, that number drops to 17%.
I don't read that as a failure by teachers. I've watched my parents adapt to every change thrown at them. Claudia Romero, a doctor of education who specializes in school improvement, put it well in a La Nación piece on Argentine universities published in September 2026: teaching is changing slowly because professors' capacity to teach with AI is very uneven. Teachers don't need another policy memo. They need time, feedback, and people to figure this out with.
One caveat on all of this. Almost every study above comes from higher education, because that is where the research is concentrated right now. The pressure is the same in a K-12 building, where one instructional coach may support dozens of teachers, and where the support gap is older than AI.
The recommendations converge, and they are specific. Look for learning where it actually happens: in discussion, in the questions students ask, in thinking you can watch unfold. MIT's own recommendations run that way: oral exams, long-term projects, portfolios, and more in-person collaboration. Use AI as something that asks questions, not something that hands out answers. That framing is Guadalupe Dorna's, who directs the public policy master's at Universidad Torcuato Di Tella: if a student is using AI to learn, the most useful answer is the one that asks a question back. Rebuild the human connections AI quietly replaced.
All of that means changing how we teach. And nobody changes how they teach from a memo. It takes reflection, honest feedback, and colleagues who can see what you can't.
Full disclosure: I work at Sibme. Here is why I think it belongs in this conversation.
Teaching becomes visible. Educators record a lesson or a presentation and review it with time-stamped comments, on their own or with colleagues. When the final product can't prove learning, watching the process can.
Reflection gets specific. A teacher finishes a lesson and can see how much of the talk in the room was theirs, and what kind of questions they actually asked. Sibme AI does the transcription and the analysis underneath. That's the skill this moment asks for: asking instead of answering. And it's AI helping people think, not thinking for them.
Nobody figures it out alone. Huddles give educators and coaches a shared space to connect, exchange feedback, and try new approaches together. It's the opposite of the isolation MIT describes.
AI raised the bar for what education has to be. I don't think the answer is better detectors. I think it's giving educators, people like my parents, the time and support to grow together.
If your school or district is working through this right now, I would rather hear what you are trying than tell you what we do.
Sources
EFE (via SWI swissinfo.ch), "Mayor universidad de México hará examen presencial a seleccionados tras anomalías con IA," July 31, 2026.
.jpg)
While AI adoption is widespread in education, relying on AI detectors is failing and fueling an atmosphere of distrust. The deeper threat is "cognitive surrender"—where students lean on chatbots at the first sign of struggle—combined with a massive support gap for teachers who lack adequate guidance. Rather than policing students, the path forward involves focusing on active, observable learning processes, encouraging human connection, and giving educators collaborative tools like Sibme to reflect on and improve their teaching practice.
A university researcher's new argument about what gets lost when AI takes over parts of teaching turns out to be a preview of a debate we're already having about teacher coaching.
Schools can't just teach kids to understand AI — they have to teach kids to recognize their own agency in the relationship with it, because literacy alone won't settle the question this week's AI-safety debate couldn't.
The instructional evidence platform for K–12 districts.
