This week
in 30 seconds
Four things worth knowing before your next department meeting.
UK undergraduates using AI on assessed work, per the HEPI/Kortext 2026 survey — up from 88% a year ago. A widely shared HEPI essay argues universities should certify "verifiable judgment," not chase detection.
OpenAI courts campuses: $100 Codex credits for verified U.S. and Canadian university students, plus a ChatGPT Edu rollout at the University of Phoenix.
Brown opened a formal investigation into the AI-cheating allegations covered last issue — and asked the professor to file individual complaints, a response he calls "appalling and insufficient."
States are racing to write AI-in-education rules: FutureEd's updated tracker shows several routing the work through their public universities, including Hawaii and Maryland.
Main stories
The developments most likely to reach your classroom, your committee, or your inbox.
94% of UK undergraduates now use AI on assessed work
The HEPI/Kortext 2026 Student Generative AI Survey found 94% of UK undergraduates use AI for assessed work, up from 88% a year earlier — overall use is 95%, and the share admitting to submitting AI-generated text directly has climbed from 3% in 2024 to 12%. A HEPI essay by dean Mauricio Villena (Diego Portales University, Chile) argues the answer is not detection but guaranteeing a minimum share of "demonstrably non-delegable" assessment: oral defenses, live problem-solving, supervised tasks.
The same week, a framework in Frontiers in Artificial Intelligence (July 18) proposed four pillars of "AI-resilient assessment" — process documentation, oral defense, authentic task design, and transparent AI-use policies — while the University of Sydney's "two-lane" model keeps surfacing as the practical template, echoing the University of Bath move covered last issue.
Read the HEPI essay →The UK numbers are a preview for every campus. Pick which of your assessments must be verifiable in person and redesign those first — the Frontiers framework gives you a defensible structure to bring to your curriculum committee.
OpenAI courts campuses: $100 student credits and a University of Phoenix deal
OpenAI is offering verified U.S. and Canadian university students $100 in Codex credits (SheerID verification, no paid plan required), pitched at course-registration agents, study tools, and startup prototypes. The University of Phoenix will deploy ChatGPT Edu institution-wide and says it will measure AI-enabled learning outcomes among working-adult students in real time.
For your campus: students arrive in the fall with subsidized frontier tools; policies written with "free ChatGPT" in mind understate what's in their hands.
Offer terms →Brown opens investigation as cheating story goes national
New since last issue: Brown formally opened an investigation (penalties range from reprimand to expulsion), the story hit the Washington Post and Boston Globe, and 27 of 86 students dropped after the final moved in person — 22 had scored 100 on the midterm. Brown asked Prof. Serrano to file individual complaints per student; he calls the response "appalling and insufficient." Canadian institutions report parallel cases (CBC).
Research
& practice
What the studies say, and what colleagues are actually doing about it.
From the Research
Deeper student–AI "collaboration" doesn't automatically mean deeper learning
A Frontiers in Psychology study found motivated learners benefit from AI as a self-directed support tool — but deeper collaborative interaction did not produce stronger knowledge application, likely because students offload analysis and overestimate their competence. Structure AI-permitted tasks so students remain the analyst: verify, critique, and correct AI output rather than co-produce with it.
Source →GenAI in lab courses needs explicit boundaries
A qualitative case study of generative AI in advanced physics laboratories found productive use required explicit design boundaries, guidance on legitimate versus prohibited practices, and verification routines that preserve students' responsibility for knowing why results are true. For lab courses, write a one-paragraph AI boundary for each phase: design, data analysis, write-up.
arXiv:2607.11417 →How Others Are Doing It
St. Lawrence College: from Copilot demos to working agents
A second full-day bootcamp (Microsoft Learning for Educators program) had staff build custom Copilot agents for meeting summaries, Excel analysis, and email drafting — under approved AI guidelines with explicit data-handling guardrails. Borrow: pair tool access with written guidelines and hands-on training in narrow, task-specific agents; access alone is not a strategy.
Source →A business professor making AI instruction explicit
C.C. Chapman committed his fall courses to explicitly teaching AI-assisted brainstorming, ideation, and planning — "we don't hand nursing students needles without instruction" — citing Microsoft's finding that 79% of students say using AI effectively matters to their future success. Borrow: dedicate one early-semester session to modeling strong versus lazy AI use in your discipline.
Source →Quick hits
& radar
Seven more stories worth a minute, plus dates and one thing to try.
-
1Does your college need an "AI librarian"? Chronicle · Jul 15 · Dartmouth & others hire librarians to lead AI literacy
-
2Now Democrats are losing confidence in higher ed, too Chronicle · Jul 16 · 46% say AI will make degrees less important
-
3FutureEd's 2026 state AI-in-education legislative tracker updated Jul 13 · Hawaii's task force sits inside the University of Hawaii
-
4Ethan Mollick named to the TIME100 Creators 2026 list TIME · Jul 14 · The Wharton professor's AI-and-education reach, recognized
-
5Vietnam leads Southeast Asia in using Google AI for education VnExpress · Jul 15 · 17% of prompts education-related; 160K students monthly
-
6Zoho launches Classes 2.0 Indian Express · Jul 15 · LMS with AI grading and a subject-restricted tutor
-
7"You can't outsource understanding" Forbes · Jul 19 · Kentucky & UW-Milwaukee teach hands-on AI judgment at scale