For University Faculty & Staff · ~10 min read · May 11–17, 2026 (ET) Newsletter Home
Intelligence Briefing

AI in
Education

The week the integrity machinery cracked: Princeton retired a 133-year honor-code tradition, arXiv started banning AI-slop authors, and the EU warned that grading with ChatGPT may already be unlawful.

133 yrs
Princeton unproctored-exam tradition, now ended
~30%
Princeton seniors who admitted to cheating
1-year
arXiv ban for unchecked AI in preprints
$200M
Anthropic–Gates AI health & education pledge
Issue briefing video

Watch the Issue 2 briefing.

A short video companion to the second AI in Education issue, placed here before the section navigation so readers can watch before diving into the full briefing.

If you read nothing else,
read these five.

The biggest moves in AI and higher education, May 11–17, 2026.

🎓 Higher Ed · The Lead Story
133

Princeton ended 133 years of unproctored exams. Faculty voted May 12 to deploy proctors at all in-person exams from July 1, after a campus survey found roughly 30% of seniors admitted cheating and the Honor Committee said AI had made peer reporting unworkable.

📊 Research
1 yr

arXiv will ban authors whose preprints show unchecked LLM output — hallucinated references, leftover chatbot meta-comments — for a full year, with peer review required to return.

🏛️ Policy · 🌍 Global
2027

Grading with ChatGPT may breach the EU AI Act. A policy expert warns informal AI assessment is likely "high-risk"; those rules apply from 2 Dec 2027, and the EU Council just called for a human-centred approach.

🎓 Higher Ed
50%

Half of campus CTOs question AI's ROI. Inside Higher Ed's 2026 survey found returns are unclear or below expectations even as adoption climbs — budget scrutiny is coming.

🛠️ Tools · 🌍 Global
$200M

Anthropic and the Gates Foundation pledged $200M over four years for AI in health and education, while Google opened free AI literacy training to all 6M U.S. teachers.

Four developments
worth your attention.

Reported and corroborated against primary publishers the week of May 11–17, 2026.

Princeton ends its 133-year unproctored-exam tradition over AI cheating

Princeton faculty voted on May 12 to deploy proctors at all in-person exams beginning summer term — retiring an Honor Code tradition in place since 1893. Dean of the College Michael Gordin pointed to a Daily Princetonian survey in which roughly 30% of seniors admitted cheating and about 45% said they knew of a violation they did not report.

The details
  • Effective July 1, 2026; faculty observe and document but do not intervene mid-exam — acting as "an additional witness."
  • Multiple departments are also reverting to handwritten blue books and oral exams.
  • Honor Committee chair Nadia Makuc said the change relieves students pressured to accuse classmates suspected of AI use.
  • Smartphones and discreet devices, Gordin noted, have made misconduct far harder for peers to spot or report.
Read the full report →
What this means for your campus

Peer-witness honor systems are visibly buckling under AI. Expect renewed pressure to either re-proctor exams or redesign assessments — and ask now whether your current assessments would even survive proctoring, or need rebuilding first.

arXiv will ban authors for a year over unchecked AI in preprints

arXiv computer-science chair Thomas Dietterich announced that any submission with "incontrovertible evidence" of unchecked LLM output — hallucinated references, leftover chatbot meta-comments — triggers a one-year ban for every listed author. After the ban, future submissions must first be accepted at a peer-reviewed venue.

Worth noting Detection is contested — meta-analyses suggest LLM-writing detectors are unreliable, so the policy targets only obvious authoring failures, not "AI-assisted" drafts.
What this means for your campus PhD students you supervise now carry career risk for every uncited or hallucinated reference an LLM produces. Spot-check before submission.
The Verge →

EU warns: grading with ChatGPT may already breach the AI Act

European University Association policy lead Thomas Jørgensen told Times Higher Education that academics quietly using ChatGPT to grade student work may already fall under the AI Act's high-risk regime — without oversight or compliance documentation. High-risk education rules apply from 2 December 2027, and the EU Council adopted conclusions May 11 calling for a human-centred, teacher-led approach.

What this means for your campus Informal "drop it into ChatGPT to mark it" practice now carries regulatory exposure — at minimum it needs an institutional policy and disclosure.
Times Higher Ed →

Half of campus tech leaders now question AI's return on investment

Inside Higher Ed's annual CTO survey found roughly half of chief technology officers say the return on AI investment is unclear or below expectations — even as adoption keeps climbing. EDUCAUSE senior director Mark McCormack warned that campuses "can no longer afford impulsive or quick-fix purchases." Rutgers' Britt Paris, who chairs the AAUP's AI committee, told IHE the muted signal "suggests that educators, students and faculty aren't finding AI to be all that useful" for teaching itself — while administrative time-savings partly reflect doing too much with too few staff. The findings landed alongside the Canvas LMS ransomware incident, which has CTOs reassessing third-party AI vendor risk.

What this means for your campus Budget conversations this summer will increasingly demand defined outcomes — not pilots — before any further AI procurement.
Inside Higher Ed →

Evidence from the field
and how peers are using it.

New studies distilled to the takeaway — plus what real instructors tried this week.

From the Research

Assessment · Berkeley CSHE

Does AI accelerate grade inflation? A 500,000-grade study

What was studied: A new Berkeley Center for Studies in Higher Education working paper (May 13) uses a difference-in-differences design on more than 500,000 grades at a large research university, 2018–2025, to estimate generative AI's effect on grade distributions and on the signalling value of grades.

What they found: Generative AI is associated with measurable upward shifts in grades for AI-exposed assignment types — degrading their value as signals for graduate admissions and hiring.

What you could do with this: Treat assignments where AI can substitute for student effort as unreliable signals; pair them with in-person or process-based evidence before writing recommendations or ranking applicants.

Working paper →
Governance · Frontiers in Education

How institutions actually govern generative AI — a PRISMA review

What was studied: A Frontiers in Education systematic review (May 12) following the PRISMA 2020 protocol, clustering institutional approaches into governance, implementation, outcomes, efficiency, faculty adoption, and ethics.

What they found: Governance is mostly top-down and policy-first, but learning-outcome gains track most strongly with institutions that pair clear policy with faculty development and assignment redesign — not with detection-tool spending.

What you could do with this: If your campus is weighing where the next AI dollar goes, the evidence base points to faculty development over additional detection licences.

Read the review →

How Others Are Doing It

Health Professions · Stanford

A scalable AI patient for clinical reasoning, across 21 universities

Who: Dr. Marcos Rojas, a physician and Stanford Graduate School of Education PhD candidate, who directs Clinical Mind AI.

What they did: Built a custom GPT that lets health-professions instructors author multilingual, interactive clinical simulations in minutes rather than weeks — now deployed at 21 universities.

What you could borrow: The workflow translates to any case-based discipline — law, social work, business ethics — letting you spin up adaptive role-plays from existing materials without building a tool.

OpenAI for Education →
Course Policy · STEM Faculty

"Show your prompt" — a one-line fix that beats detection

Who: A STEM faculty member writing in Inside Higher Ed (May 14), arguing that blanket prohibition is the wrong response to AI.

What they did: Added one sentence to the course AI policy: "If you use AI, submit the prompt and the output alongside your work."

What happened: Students who outsourced their thinking self-exposed; thoughtful users demonstrated their reasoning — with no detection software in the loop.

Inside Higher Ed →
Try This Week — three things, under 30 minutes each
1

Add one syllabus line: "If you use AI, submit the prompt and the output with your work." (20 min)

2

Spot-check three reference lists for hallucinated citations before any preprint or thesis chapter goes up. (25 min)

3

Tag every assessment in one course on the Perkins 5-category AI-use scale using a free template. (30 min)

Everything else
worth a scan.

Eight stories in under a minute — plus dates and questions to take to your department.

On My Radar — Dates
July 1, 2026
Princeton's proctored-exam policy takes effect for summer term.
June 28–July 1, 2026
ISTE 2026, Orlando — Turnitin demos its Google Classroom integration (booth 1158).
Dec 2, 2027
EU AI Act high-risk education provisions begin applying; Commission guidance expected before then.
Talk About It

Princeton re-proctored exams after 133 years. If your department took the same vote next week — would your current assessments survive proctoring, or need redesign first?

arXiv now treats unchecked LLM output as the author's misconduct. Should your department apply that same standard to grant proposals, thesis chapters, and recommendation letters?

Try This Week

Add one line to your syllabus: "If you use AI, submit the prompt and the output with your work." It converts AI from a policing problem into evidence of process — and takes under 20 minutes.