Most institutions are writing or planning a policy about AI; few have changed what they actually measure. The essay, the problem set and the report were proxies for thinking that worked only while producing them was expensive. That assumption is gone. Detection tools are unreliable enough to risk accusing the wrong students, and much of what is graded no longer shows reliably what a student understands.
Who lives with itTeachers and lecturers redesigning assessment alone · students wrongly suspected · institutions whose credentials depend on the answer
Why nowA UNESCO survey found 19% of higher education institutions had formal AI guidance with a further 42% developing it, while more than half reported uncertainty about how to apply AI pedagogically and one in four had already encountered ethical problems including authorship disputes.
Why it mattersThe question has moved from catching misuse to evidencing learning, and nothing is built for the new question. Making a student's reasoning visible, cheaply and without treating them as a suspect, solves what detection never could.
Students choosing what to study need a fair picture of the work their degree leads to. The evidence they are given describes an earlier market. Official long-run earnings figures, and the rankings that use them, follow graduates who started work before generative AI. Government job forecasts are built on historical patterns, and their authors warn against reading them as predictions. School advisers are stretched, and the people building AI contradict one another in public on what to study. Meanwhile employers are redesigning entry-level roles in AI-exposed occupations, and some students are already switching subjects because of AI, before the evidence settles. The choice comes early, costs years and often debt, and is hard to undo. It stays unsolved because the current evidence is unsettled, the trusted evidence is old, and the two reach students separately.
Who lives with itStudents choosing what to study · parents helping to pay for a degree · school counsellors and university careers services · governments, schools and universities that fund guidance
Why nowIn September 2026 a Federal Reserve governor said many recent graduates find first jobs harder to get, “possibly partly related” to AI taking on entry-level tasks. Stanford’s Digital Economy Lab finds that employment of 22-to-25-year-olds in AI-exposed occupations has fallen 19% behind less-exposed peers, a descriptive finding, while Yale’s Budget Lab finds no significant effect so far. In a Lumina–Gallup panel, 16% of US students say they have already changed major because of AI.
Why it mattersOfficial graduate earnings statistics provide valuable evidence about past outcomes, but they cannot by themselves establish which degrees will offer the strongest prospects in a changing labour market.
The US Bureau of Labor Statistics explains that its projections assume technological progress broadly in line with historical experience, while acknowledging that unusually rapid technological change could make historical relationships unreliable.
Students and families must therefore make consequential choices by interpreting evidence that varies in age, certainty and relevance to their circumstances.
Framed by Ainna editorial
A funded startup to make a dent inCritical (Ainna's editorial grade)12 sources
The same Radar runs in your context. What you see here is Problem Radar, free. Tune it to your market and it listens to the sectors, themes and geographies you set, landing every find in your portfolio, graded: that's the Team plan (coming soon). Tune it to your company and it listens to your innovation agenda and to your own people: that's Enterprise. Both are private to you: nothing they find is ever published.
