Five things worth
knowing before Monday
If you read nothing else, read this.
A Brown University economist says he has “overwhelming evidence” that more than 50 students used AI to cheat on an exam — the largest known AI-cheating case in the Ivy League — and is demanding a public reckoning rather than quiet case-by-case handling.
A 26,811-student study found generative AI lifted homework scores 18% but cut closed-book exam performance ~20% — when it replaced, rather than supported, thinking.
Of 96 UK universities with public AI policies, only 4 use “detection-dominant” language — but HEPI warns many education-first policies still live in the misconduct handbook.
Google rolled out Gemini “study notebooks,” Guided Learning, and no-cost ACT/GRE practice tests at ISTE 2026 — course-grounded study tools headed for your students by default.
As detector accuracy tops out near 76%, universities are reviving in-person exams — while proprietary AI-likelihood scores leak onto resale markets.
What moved this week
— and why it matters
The handful of developments a busy professor should actually act on.
Brown economist alleges mass AI cheating — the biggest known Ivy League case
Roberto Serrano, a chaired economics professor at Brown, says he has “overwhelming evidence” that at least 50 students in his advanced course (ECON 1170) cheated using AI on a March midterm after he switched to take-home exams. He argues faculty “cannot be left on their own” and that institutions must publicly admit the scale of the problem.
For context, a Harvard analysis notes 47% of seniors admitted some cheating, and Princeton recently dropped a 133-year ban on proctoring exams, citing AI.
El País, June 28 → Fortune, June 24 →Worth notingSerrano describes the evidence as conclusive, but reporting does not detail the method; AI-use claims based on pattern inference remain contestable. A documented adjudication process and disclosed evidentiary standard would resolve it.
Expect renewed pressure to define — in writing and before the term starts — what counts as AI misuse and what evidence your department will actually act on.
HEPI: AI policies talk “education” but often live in the misconduct handbook
Of the UK’s 163 universities, 96 had public AI policies; 83 used education-dominant language and only 4 were detection-dominant. But HEPI Policy Note 71 finds that policies filed under misconduct “inherit that framework’s assumptions about students” — where guidance sits may matter as much as what it says.
HEPI → THE →26,811-student study: AI lifts homework scores but lowers later exams
A CEPR paper (“The Generative AI Learning Penalty”) tracked Chinese secondary students over 30 months: AI cut homework time ~30% and raised homework scores 18%, while closed-book exams fell ~20% within six months. 80% of the loss came from students who finished fast — outsourcing the thinking. Ethan Mollick: “AI tutoring in support of classes is good; using AI to ‘help’ with homework is bad.”
Dataconomy → Psychology Today →Google ships free Gemini study tools and ACT/GRE practice tests
At ISTE 2026, Google for Education launched “study notebooks” (personalized lessons, quizzes, study guides) and “Guided Learning,” plus free full-length ACT and GRE practice tests with The Princeton Review, slated for summer 2026. Florida State University is highlighted putting NotebookLM in students’ hands.
Google → EdTech Innovation Hub →The detection backlash: in-person exams return, scores leak to resale markets
As detector confidence erodes (most top out near 76% accuracy), Cardiff, Durham and Edinburgh Napier now require AI-use “declaration” statements, and many institutions are reweighting toward proctored exams. THE warns universities have “inadvertently” created an illicit detection economy where likelihood scores are sold and gamed. Australia saw thousands of misconduct accusations later dismissed.
THE Opinion → Policy review →The evidence —
and how peers are using it
What the studies say, and what colleagues are actually trying in their classrooms.
From the Research
Open-book vs. closed-book: does AI framing change learning?
A graduate biostatistics course measured AI use against weekly open-book homework and weekly closed-book in-class quizzes. Students were briefed up front on “good” AI use (explaining concepts, generating practice) vs. “misuse” (producing final answers) — a real-world test of whether framing changes behavior. [Preprint; not yet peer-reviewed.]
Do this: pair every AI-permitted homework with a short closed-book check on the same skill; the gap reveals whether learning transferred.
Research Square →73% of faculty have handled an AI integrity case
A January 2026 AAC&U survey found 73% of faculty reported personally handling an AI-related academic-integrity issue, and 83% predicted AI will shorten students’ attention spans. [Secondary source citing AAC&U; verify against the original before quoting in policy.]
Do this: use the 73% figure to argue integrity workload is a shared faculty-development problem, not one instructor’s.
Statistics roundup →How Others Are Doing It
Make critiquing the AI the graded task
Who: Prof. Anders, a writing instructor whose course was studied and profiled in The Chronicle (June 29). What they did: students co-wrote with chatbots and were required to find and correct the tools’ mistakes, then refine prompts. What happened: students were more reflective and purpose-driven — judging output quality was itself “pretty deep learning.”
Borrow this: make fixing a flawed AI draft the assignment, not generating one. Thinking shifts from production to evaluation.
Chronicle →From substitute to support: reframe AI as a writing assistant
Who: a writing instructor in Faculty Focus (June 26). What they did: reframed AI in the syllabus as a writing assistant, not a substitute, with guided practice and reflection. What happened: students engaged more actively with revision; “clear expectations and space to reflect” beat blanket bans.
Borrow this: replace one “don’t use AI” clause with a specific “here’s how to use AI on this assignment” note.
Faculty Focus →Scan it in a minute,
plan the weeks ahead
Everything else worth knowing, plus what’s coming.
- 1Harvard cheating data; Princeton ends proctoring ban47% of Harvard seniors admit some cheating as Princeton drops a 133-year-old ban, citing AI.
- 2Google’s AI talent exodus continuesTwo more Gemini researchers reportedly leave for Anthropic — shaping the tools students will use.
- 3EDUCAUSE: a campus AI awareness gap94% of higher-ed workers used AI in six months; only 54% knew their institution’s policy.
- 4AAC&U / Elon ‘Student Guide to AI’ keeps spreading35,000+ downloads across ~4,000 institutions in 170 countries; free, CC-licensed.
- 5The Atlantic: AI tutors won’t fix motivationA Stanford review of K-12 evidence finds generally limited learning benefits.
- 6University of Idaho launches three AI degreesNew programs plus NAIRR access — investing in building AI talent, not just policing it.
- 7HEPI/Kortext: 94% of UK undergrads use AI on assessed workTwo-thirds say assessment has changed significantly — adoption has outrun policy.