aiX | TZStats aiX Lab | Tian Zheng, Columbia Stats

aiX Weekly — AI in Higher Education (July 29th, 2026)

Wed, 29 Jul 2026 00:00:00 +0000

This week’s post centers on three threads: the durability of learning — whether learners’ gains through AI use survive once the tool is removed; assessment as the load-bearing response to that problem; and the redefinition of entry-level work as AI skills become a hiring baseline.

Each aiX Weekly post is organized around a set of recurring sections and pairs with our companion AI and Higher Education timeline, which traces the broader arc of how AI has reshaped higher education since late 2022.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

Reviewed by Tian Zheng on July 29th, 2026.

Scope note: this issue covers developments from roughly July 15–22, 2026, organized around three threads — the durability of learning, assessment as the load-bearing response, and the redefinition of entry-level work. Research Highlights also cites one earlier conceptual reference — a 2024 Nature Perspective — as grounding for this week’s empirical findings.

TL;DR

Durability of learning: A large study of 26,811 secondary students found AI raised homework scores 18% and cut homework time 30% — then, within six months, monthly exam scores fell 20% and college-entrance scores fell 18–24%, with most of the decline among students who “outsourced” work to AI (Research Highlights)
The same worry, one level up: A 2024 Nature perspective warns that AI can create “illusions of understanding” and narrow inquiry into “scientific monocultures” — the research-level analogue of “produce more, understand less” (Research Highlights)
Assessment as the load-bearing wall: An EDUCAUSE report and coverage of unreliable detection tools both point the same way — toward redesigned, process-visible, or in-class assessment rather than surveillance (Research Highlights · What’s in the News)
Workforce redefinition: Demand for AI skills in entry-level jobs has roughly tripled since fall 2025 even as junior work shifts toward judgment, and Purdue makes an “AI working competency” a graduation requirement this fall (What’s in the News)

This Week at a Glance
Research Highlights
What’s in the News
Most Discussed
Try This Week

This Week at a Glance

Three threads organize this issue. The first is durability of learning — whether a learner’s gains through using AI survive once the tool is removed: a large secondary-school study finds homework gains reversing into medium-run exam declines, and a Nature perspective raises the same “produce more, understand less” worry at the level of research. The second is assessment as the load-bearing response — if AI-assisted work no longer indexes learning, the unaided, observed task is where a grade recovers its meaning, a shift visible in the EDUCAUSE assessment report and the retreat from detection tools. The third is workforce redefinition — entry-level roles increasingly assume AI skills even as junior work tilts toward judgment, and institutions such as Purdue are writing AI competency into the degree.

Relevant to faculty: The homework-vs-exam gap is a measurement handle — a divergence between AI-assisted and unaided performance is observable, and assignments can be designed to surface it early, definitely before the exams.

Relevant to institutional leaders: The workforce signals — tripling AI-skill demand, competency requirements — raise a curriculum question: where do graduates build judgment if the routine tasks that once developed it are increasingly automated?

Relevant to students and researchers: This week’s strongest study measured learning months later, not at submission — a reminder that immediate performance and durable knowledge can move in opposite directions, whether the work is a homework set or a research paper.

Research Highlights

Faster Completion, Less Learning: Generative AI Reduced Study Time and the Knowledge Students Build (CEPR Discussion Paper 21577; coverage in Fortune, July 21) Researchers from Stockholm University and the University of Hong Kong tracked 26,811 students in grades 7–12 and found that AI adoption raised homework scores by 18% and cut homework completion time by about 30%. Within six months, however, monthly exam scores fell by 20%, and college-entrance-exam performance fell by 18–24%, with the worst results appearing about two years after adoption. The authors attribute roughly 80% of the decline to students who “outsourced” homework — using AI to complete tasks accurately but quickly — rather than to learn from them.

Large-scale panel / quasi-experimental

💬 Editor’s note: This is the mirror image of last week’s proctored experiment: there, students who used AI to understand kept their gains; here, students who used it to finish lost ground months later. The consistent signal across very different designs and populations is that the value shows up in unaided, delayed measurement — which is exactly where most course assessment doesn’t look. The 80%-of-decline-from-“outsourcers” figure is worth confirming in the primary paper before citing. — TZ

Uncovering the Multidimensional Effects of Generative AI on Learning (Sungkyunkwan University; via Phys.org, July) A team at Sungkyunkwan University ran an experiment with 88 university students that split learning into three stages — concept understanding, problem solving, and result review — and randomly assigned AI (GPT-4o) support to different stages. AI help during the problem-solving stage improved test performance relative to AI limited to concept understanding, which the authors attribute to reduced cognitive load; they also flag concerns about dependence when AI is available throughout. (Secondary coverage; the primary paper and effect sizes should be verified before citing specifics.)

Stage-randomized experiment

💬 Editor’s note: The interesting claim is that where in the learning cycle AI enters may matter as much as whether it’s present — support at problem-solving helped more than at first exposure. That’s a testable design principle. — TZ

The Impact of AI on Learning Assessment (EDUCAUSE, June 2026 report) This EDUCAUSE report surveys how institutions are rethinking assessment as generative AI makes many take-home tasks trivial to complete, documenting movement toward authentic, process-visible, and oral or in-class formats and away from reliance on AI-detection tools. It positions assessment redesign — rather than detection — as the primary institutional response. (Report; included for direct relevance to this week’s assessment thread. Detailed findings should be read in the source.)

Institutional survey/report

💬 Editor’s note: Placed alongside the two studies above, the throughline is that assessment is becoming the load-bearing wall: if AI-assisted homework no longer indexes learning, the unaided, observed task is where a grade recovers its meaning. This is a measurement problem before it is a policy problem. — TZ

Artificial Intelligence and Illusions of Understanding in Scientific Research (Messeri & Crockett, Nature Perspective, March 2024) This 2024 Nature Perspective develops a taxonomy of how scientists envision AI across the research pipeline and argues that its appeal rests on promises to raise productivity and objectivity by compensating for human limits. The authors caution that the same tools can produce “illusions of understanding” — a sense of comprehending more than one actually does — and can foster “scientific monocultures” in which a narrow set of methods, questions, and viewpoints crowds out alternatives, leaving a science that produces more while understanding less.

Perspective / conceptual framework

Editor’s note: Set next to this week’s empirical items, this article is highly relevant today: “produce more but understand less” is the research-level analogue of the homework-vs-exam gap in student learning. — TZ

What’s in the News

Time for Class 2026 Finds Administrators Now Out-Use Students on AI (Tyton Partners, released June 15; coverage this week) | D2L summary Tyton Partners’ Time for Class 2026, subtitled “The AI Tipping Point,” reports that more than half of students, faculty, and administrators now use AI at least weekly, and that administrators use AI daily at a higher rate (43%) than students (32%) — a reversal after years of students leading. Only 22% of faculty consider their institution’s AI policy effective, versus 44% of administrators.

Demand for AI Skills in Entry-Level Jobs Nearly Triples Since Fall 2025 (NACE) NACE reports that the share of entry-level postings requiring AI skills has risen sharply — by some measures roughly tripling since fall 2025 — even as overall class-of-2026 hiring is projected up about 5.6%. Roughly 35% of entry-level roles now reference AI skills, and employers describe junior work shifting toward more analytical and judgment-based tasks and away from routine ones.

💬 Editor’s note: If foundational tasks are thinning while judgment tasks grow, the pipeline question is where new graduates build judgment if the routine reps that used to develop it are automated. That’s a curriculum question as much as a labor-market one. — TZ

Purdue’s “AI Working Competency” Graduation Requirement Goes Live This Fall (Washington Times, July 9) | Purdue newsroom Purdue confirmed that its “AI working competency” graduation requirement takes effect with students entering this fall, part of the broader AI@Purdue strategy spanning learning with AI, learning about AI, research, and operations. It makes a discipline-general AI competency a condition of the degree rather than an elective or a career-services add-on. (Previously covered in Issue #5 — the trustees’ approval; this week added implementation timing.)

💬 Editor’s note: This sits in the workforce-redefinition thread alongside the NACE data above: if entry-level roles increasingly assume AI skills, writing competency into the degree is one institutional answer. The part I’d be very curious to know more about is how “competency” gets assessed. — TZ

Most Discussed

The Atlantic’s “Post-Literate Age” Cover Story A widely-shared Atlantic cover story argues that a shift toward fast, convenient information intake — accelerated by AI — is eroding sustained reading and, with it, the capacity for critical thought. It circulated alongside this week’s homework-vs-exam study as a cultural counterpart to the empirical finding.

💬 Editor’s note: Read as argument rather than evidence, it names the mechanism the CEPR study quantifies: when the effortful part of learning is skipped, the capacity that effort built can erode. The empirical work suggests the outcome tracks how the tool is used, which is a more actionable claim than generational decline. — TZ

“We Have Never Taught Critical Thinking” (Inside Higher Ed, opinion, June 23) This opinion piece, still circulating in faculty discussions, pushes back on the “AI is killing critical thinking” framing by arguing that higher ed has rarely taught critical thinking explicitly in the first place — and that AI is exposing, not causing, a pre-existing gap. It reframes the panic as an opening to be more deliberate about what critical thinking means and how it’s assessed.

💬 Editor’s note: Exposing a pre-existing gap is also an opportunity calling for action. Statistical thinking is a form of critical thinking that could be very valuable to anyone in the age of AI. — TZ

“The Teaching Machine, Again”: Historical Skepticism Resurfaces (via Fortune, July 21) Alongside the homework-vs-exam study, commentators revived a century-old comparison: Pressey’s 1924 and Skinner’s 1950s “teaching machines,” which produced strong performance while students used the device but failed to transfer once it was removed. Neuroscientist Jared Cooney Horvath frames AI as a possible repeat of that “transfer problem” — tools experts use to save effort may teach novices dependency rather than skill.

Try This Week

The Delayed, Unaided Check (20–30 minutes)

Inspired by this week’s studies, where AI-assisted gains faded on later, unaided measures:

Pick one assignment where students may use AI. Note what a strong submission currently demonstrates.
Add a brief, unaided follow-up a week later — three minutes of in-class writing, a short oral question, or a quick problem — that asks students to reuse the same idea without AI.
Compare the two, informally. A large gap between the assisted and unaided work is the signal this week’s research would predict for “outsourced” learning.
Consider sharing the rationale with students: the follow-up isn’t a trap but a way to confirm the learning transferred — which is the outcome that persists after the tool is gone.

The aim is not to catch AI use but to make durable understanding visible, so both you and the student can see whether it took hold.

aiX Weekly is curated by Claude and reviewed by Tian Zheng for the aiX Faculty Fellowship at Columbia University. Editorial notes reflect one statistician’s reading of the week, offered to prompt discussion rather than to prescribe. Corrections and suggestions welcome.

aiX Weekly — AI in Higher Education (July 22nd, 2026)

Wed, 22 Jul 2026 00:00:00 +0000

This week’s post examines two converging disruptions: accountability when AI mediates consequential decisions, and AI’s growing strain on the knowledge and talent ecosystems — peer review, open-source software — that it was built on.

Each aiX Weekly issue is organized around a set of recurring sections and pairs with our companion AI and Higher Education timeline, which traces the broader arc of how AI has reshaped higher education since late 2022.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

Reviewed by Tian Zheng on July 21st, 2026.

TL;DR

A human-subjects study finds that reviewers told to verify AI output grew more confident in the system’s answers — even when wrong — while a pre-registered experiment shows structured oversight can reduce AI research failures from 72% to 16% (Research Highlights)
AI is disrupting the peer review system it was built on: journal submissions up 42% since ChatGPT, 21% of ICLR 2026 reviews fully AI-generated, and an Organization Science study finds AI-generated reviews are measurably narrower and lower quality (Research Highlights)
Twenty-six Meta employees sue over AI-driven layoff selections, alleging algorithmic performance systems made the actual decisions while humans rubber-stamped the output (Most Discussed)
Princeton’s Arvind Narayanan argues at ICML 2026 that AI is a “normal technology” — powerful but unreliable — and warns that automating peer review is “a trap” that cedes control of research direction (Most Discussed)

This Week at a Glance
Research Highlights
Institutional Movements
What’s in the News
Most Discussed
Interesting Ideas & Repos
What Changed

This Week at a Glance

Two disruptions converge in this week’s material. The first is to accountability: when AI sits between a decision-maker and a consequence — scoring employees, reviewing research, verifying student work — who owns the judgment? The confirmation-bias paper shows individual verification can backfire. The “(Human) Attention” paper shows structured oversight works when designed as architecture. Meta’s lawsuit puts the question in legal terms.

The second disruption is to the knowledge and talent ecosystem itself. AI is straining the very systems that produced it. The peer review process that validates research is being flooded with AI-generated submissions and degraded by AI-generated reviews. The open-source software ecosystem that AI’s infrastructure depends on is being overwhelmed by AI-generated contributions that consume volunteer review time. And Narayanan’s ICML keynote argues that if we automate the evaluation layer — peer review, human oversight, quality control — we cede the steering of research to the systems we’re supposed to be directing.

Both disruptions share a common structure: AI is powerful enough to increase volume but not reliable enough to maintain quality, and the gap between the two is being absorbed by the human systems — reviewers, maintainers, faculty, students — that were already stretched thin.

Relevant to faculty: The peer review and open-source stories are not someone else’s problem. Faculty depend on both systems — for the literature they teach from and the tools they build with. The confirmation-bias and “(Human) Attention” papers together suggest that “verify the output” needs structure, not just instruction.

Relevant to institutional leaders: The DEC’s 6% faculty-support figure, and Handshake’s 28% integration gap all point to an accountability distribution problem. The peer review crisis adds a research-infrastructure dimension: the quality of the knowledge pipeline institutions depend on is itself under pressure.

Relevant to students and researchers: Narayanan’s “normal technology” framing — and his warning that automating peer review is “a trap” — is directly relevant to anyone entering a research field. The Meta lawsuit illustrates the same questions about algorithmic scoring and nominal human review that show up wherever data-driven systems shape outcomes.

Research Highlights

Confirmation Bias: A Challenge for Scalable Oversight Across two human-subjects studies of simple oversight protocols, participants told that a model is “correct most of the time, but not all of the time” became more confident in the system’s answers after conducting their own online research — even when those answers were incorrect. Showing arguments for both candidate answers improved accuracy in the cases where the model was wrong.

Study type: human-subjects experiments

Editor’s note: This complicates the default pedagogical move of “have students check the AI.” If the act of verifying can increase confidence in wrong answers, then oversight training may need to teach structured disconfirmation — actively seeking the case against — rather than open-ended checking. — TZ

(Human) Attention Is (Still) All You Need: Human Oversight Makes AI-Assisted Social Science Reliable In a pre-registered 2×4 factorial experiment with 280 complete research runs across four datasets, researchers tested whether structured human oversight can make AI-assisted economic research reliable. An unconstrained multi-agent baseline produced critical failures — specification errors, hallucinated findings, unsupported conclusions — in 72% of runs. Their Human-in-the-Loop Economic Research (HLER) architecture, based on pre-commitment, decision sequencing, accountability, and attention allocation, reduced the failure rate to 16%.

Pre-registered factorial experiment

Editor’s note: Oversight may work when it’s designed as an architecture rather than an afterthought. For teaching, the four design principles (pre-commitment, sequencing, accountability, attention allocation) are transferable to how students interact with AI in research assignments. — TZ

Modeling Human Beliefs about AI Behavior for Scalable Oversight This paper addresses a foundational challenge for AI oversight: human evaluators may form incorrect beliefs about what AI systems are actually doing in complex tasks, leading to unreliable feedback. The authors formalize how evaluator belief models interact with value learning and introduce “belief model covering” as a way to reduce dependence on precise belief models. Published in Transactions on Machine Learning Research.

Theoretical/formal analysis

Editor’s note: This connects to the confirmation-bias paper earlier in this issue — both identify the same vulnerability from different angles. That paper shows evaluators growing more confident in wrong answers; this one formalizes why: their mental models of what the AI is doing can be systematically wrong. For teaching, it suggests that “check the AI’s work” is insufficient without first helping students build accurate models of what the AI actually does. — TZ

More Versus Better: Artificial Intelligence, Incentives, and the Emerging Crisis in Peer Review Journal submission volume is up 42% since ChatGPT’s release, but the review system absorbing that volume is degrading in measurable ways. This Organization Science study finds Flesch Reading Ease scores in reviews dropped 1.28 standard deviations, and AI-generated reviews are narrower — more focused on theory, less on data — than human reviews. The scale of the problem came into sharper focus at ICLR 2026, where Pangram Labs estimated 21% of 75,800 reviews (roughly 15,900) were fully AI-generated, with over half showing some AI involvement. ICML 2026 desk-rejected approximately 500 papers for LLM policy violations.

Published research

Editor’s note: When AI simultaneously floods the submission pipeline and degrades the review process, the feedback loop protecting research quality weakens at both ends. — TZ

Institutional Movements

Digital Education Council survey: 6% of faculty fully agree their institution provides sufficient AI resources In a survey of 1,681 faculty across 52 institutions in 28 countries, only 6% fully agreed that their institution had provided sufficient resources to build faculty AI literacy, even as 86% anticipated using AI in their teaching in the future. The gap points to a distance between institutional AI announcements and faculty-reported experience.

Open question: What forms of support (time, training, credit, community) actually move the 6% figure?

Handshake Class of 2026: 85% Use AI, 28% Say School Integrated It Handshake’s survey of 1,248 graduating seniors across nearly 500 institutions finds 85% used AI tools in college, but only 28% say their program “meaningfully integrated” AI. Fifty-eight percent say they’ll need stronger AI skills to succeed at work. Meanwhile, 62% of seniors report feeling pessimistic about their careers (up from 46% in 2024), with almost half citing AI’s impact as a factor. Three-quarters of students who use generative AI describe themselves as “reasonably skilled” or “very skilled” — a self-assessment that may or may not reflect professional readiness.

Open question: Is the 28% integration figure a supply-side failure (institutions not offering enough) or a recognition gap (students not identifying AI instruction they’ve received)?

What’s in the News

Inside college AI cheating wars: surveillance, false accusations, confusion (Anchorage Daily News) The report describes uneven and sometimes extreme anti-cheating practices — students asked to show their desks with mirrors during online tests, or to keep arms crossed during oral exams — alongside a rise in false accusations from probabilistic AI-detection tools. It frames the definition of “cheating” itself as unsettled.

What’s interesting here: The same detection tools meant to restore trust are generating a new category of disputed cases, which changes the burden of proof students face.

The Open-Source Maintainer Crisis: AI-Generated Contributions as Denial of Service In the first three weeks of January 2026 alone, three major open-source projects took drastic defensive measures against AI-generated contributions. curl shut its six-year bug bounty program after being flooded with AI-fabricated vulnerability reports. Ghostty implemented a zero-tolerance policy for AI-generated code submissions. tldraw auto-closed all external pull requests. Maintainers describe the situation as a “denial-of-service attack” — a flood of superficially plausible but low-quality contributions that consume review time and degrade signal. The paradox: AI companies simultaneously depend on open-source infrastructure and are undermining the volunteer labor that sustains it.

What’s interesting here: Open-source software is the foundation AI was built on — the frameworks, libraries, and training infrastructure are overwhelmingly open-source. AI is now degrading the very maintenance ecosystem it depends on.

Most Discussed

Meta Employees Sue Over AI-Driven Layoff Selections (Courthouse News Service) | CNBC Twenty-six current and former Meta employees filed a federal lawsuit alleging the company used a constellation of internal AI systems — including “Metamate” (an LLM assistant tracking internal communications), employee-trained “second-brain” agents, keystroke and activity monitoring, AI-token-usage dashboards, and algorithmically assisted performance ranking — to score, rank, and select 8,000 employees for layoffs. The plaintiffs argue these systems penalized workers on protected medical or parental leave, whose reduced digital activity produced lower scores. Meta responded that “workforce management and organizational decisions were and are made by people, not AI.” The case was filed July 13 in the Northern District of California.

Editor’s note: When performance data passes through AI systems that aggregate, weight, and rank before a human reviews the output, what exactly did the human decide? Which features were selected, how were absences weighted, what thresholds triggered a flag? These are modeling choices, made somewhere in the pipeline, by someone — or by no one in particular. This is a challenge for any data-driven process, including education. When we use AI to summarize student engagement data, flag at-risk students, or even aggregate peer evaluations, the same questions apply: what assumptions went into the score, and who owns them? — TZ

Arvind Narayanan’s ICML 2026 Keynote: “What Will Be Left for Us to Work On?” | Annotated slides Princeton’s Arvind Narayanan, in the highest-profile keynote at ICML 2026 (July 13), argued for treating AI as a “normal technology” — powerful but subject to the same decades-long adoption cycle as electricity or computing. His central claim: a “capability-reliability gap” means capability has shot upward while reliability has improved only 5–10 percentage points, so “for now, you can have only two of three: general-purpose, high-stakes, automated.” Narayanan warned against automating peer review — calling it “a trap” that cedes control of research direction — and reframed expert work as “steering rather than rowing”: effort shifts from production to evaluation.

Editor’s note: Narayanan’s “normal technology” framing is one of the most useful narratives about AI I’ve seen. His observation that code-writing was roughly a third of software engineering work and “never the bottleneck” applies directly to teaching — we need to identify which parts of learning are analogous to code-writing (automatable) and which are analogous to requirements, design, and evaluation (not). — TZ

Interesting Ideas & Repos

A Decoupled Human-in-the-Loop System for Controlled Autonomy in Agentic Workflows An architecture proposal that positions human checkpoints at specific, decoupled points in agentic workflows rather than assuming continuous oversight.

Scalable Oversight via Recursive Self-Critiquing Explores whether AI systems can critique their own outputs recursively, a candidate mechanism for “human on the loop” rather than “human in the loop.” Editor’s note: Worth pairing with this week’s confirmation-bias paper — one asks whether humans can verify, the other whether machines can help. Neither is settled. — TZ

What Changed

The disruption turned inward: AI is now straining the knowledge systems it was built on — peer review flooded with AI-generated submissions and degraded reviews, open-source maintenance overwhelmed by AI-generated contributions. This is a new category of concern, distinct from the adoption and assessment stories of Issues #1–3.
The oversight problem deepens: The confirmation-bias paper and the modeling-beliefs paper show, from different angles, that human verification itself can be unreliable — going beyond the “keep a human in the loop” framing of earlier issues.
A corrective framework arrives: Narayanan’s ICML keynote offers “normal technology” as an alternative to both hype and catastrophism, with the capability-reliability gap as its central diagnostic.
The accountability question sharpens from “should we use AI?” to “who bears the consequences when we do?” — a shift the Meta lawsuit, the peer review crisis, and the open-source maintainer crisis all make concrete.

aiX Weekly — AI in Higher Education (July 15th, 2026)

Wed, 15 Jul 2026 00:00:00 +0000

This week’s post examines the widening gap between near-universal AI adoption and institutional readiness — from Microsoft and Gallup survey data to the platform competition at ISTE 2026.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

Reviewed by Tian Zheng on July 14th, 2026.

TL;DR

Microsoft’s 2026 AI in Education Report finds 92% of students use AI for school — but 77% have received no formal training (Research Highlights)
Nearly half of college students have considered switching majors over AI career concerns — the Lumina-Gallup study documents workforce anxiety reshaping enrollment in real time (Institutional Movements)
Google and Microsoft both launched major AI education tool suites at ISTE 2026 — the platform competition for classroom AI is now explicit (What’s in the News)
Frontier AI models now score ~30 points above human PhD experts on graduate-level science benchmarks (Most Discussed)

This Week at a Glance
Research Highlights
Institutional Movements
What’s in the News
Most Discussed
Interesting Ideas & Repos
Try This Week

This Week at a Glance

This week’s through-line is the gap between adoption velocity and institutional readiness. Microsoft and Gallup data both confirm near-universal AI use in higher education, but training, policy, and assessment redesign lag far behind. ISTE 2026 saw Google and Microsoft competing to define the AI classroom platform — a vendor landscape institutions will need to navigate carefully.

Relevant to faculty: The UK HEPI study finding that most university AI policies “promise support but deliver surveillance” may resonate beyond British universities.

Relevant to institutional leaders: The Lumina-Gallup major-switching data (47% considering, 16% already switched) signals enrollment pattern shifts worth planning for.

Relevant to students and researchers: GPQA Diamond saturation — AI models scoring 30 points above PhD experts — raises questions about what knowledge assessments measure.

Research Highlights

1. Microsoft 2026 AI in Education Report: Universal Adoption, Minimal Training

Microsoft’s New AI in Education Report

Microsoft’s third annual report surveyed 3,345 respondents across K-12 and higher education in six countries, released at ISTELive 2026. The headline finding: 92% of students and 88% of educators report using AI for school-related purposes, but 77% of students and 53% of educators have received no formal AI training. Two-thirds of educators want monthly or quarterly training.

Industry-commissioned survey

💬 Editor’s note: I believe both teachers and students can benefit from AI training — training that equips them to evaluate and govern AI effectively. — TZ

2. Digital Education Council LATAM Survey: 30,000 Responses Across 29 Institutions

AI in Higher Education LATAM Survey 2026

Surveying over 30,000 respondents (22,941 students, 7,319 faculty) across 29 Latin American institutions, the study found 92% of students and 79% of faculty actively engage with AI. Notably, 65% of students worry AI may make learning “too shallow” — mirroring RAND’s US findings in a completely different cultural and linguistic context. A 31-point gap emerged between students wanting AI-assisted feedback (50%) and faculty providing it (19%).

Large-scale survey

💬 Editor’s note: Students in both the US and Latin America independently voice the same worry about AI and cognitive engagement. The feedback gap — students want more AI feedback than faculty provide — suggests a concrete opportunity for experimentation. — TZ

3. HEPI Policy Note: What UK University AI Policies Actually Do

What UK University AI Policies Actually Do: A Study of 96 Institutions

Professor Sam Illingworth computationally analyzed AI policies of 96 UK degree-awarding institutions, — 41% had no publicly accessible AI policy. Of those that did, 86% appeared education-focused by keyword count, but close reading of a subset found nearly half were actually detection-and-discipline frameworks using educational language as a veneer. Full data and coding framework available on GitHub.

Policy analysis

💬 Editor’s note: The open dataset and codes on GitHub make this replicable. — TZ

Institutional Movements

1. Lumina-Gallup: 47% of Students Have Considered Switching Majors Due to AI

College Students Weigh AI’s Impact on Majors and Careers (Gallup)

The Lumina Foundation-Gallup 2026 study surveyed 6,010 US adults who opted-in via an online panel, including 3,801 enrolled students. Nearly half (47%) have given serious consideration to switching majors because of AI, and 16% have already done so. Technology (70%) and vocational fields (71%) show the highest consideration rates. Male students (60%) are more likely to consider changes than female students (38%).

💬 Editor’s note: It’s understandable that students are reconsidering their majors in response to AI. What I’m left wondering is how students are actually making these decisions. Which factors did they weigh — projected job security, salary, how “automatable” a field seems, their own interest? And how did they arrive at those judgments? It’s still genuinely unclear which skills will hold the most value in an AI-native workforce. If students are steering away from fields on the assumption that AI will hollow them out, that assumption is doing a lot of work, and it may or may not be right. The survey captures the reaction but not the reasoning behind it. I’d want to see the reasoning before concluding these shifts are well-calibrated rather than driven by a diffuse sense of anxiety about a still-uncertain future. — TZ

2. University of Surrey: AI Embedded in Every Degree from September 2026

AI to Be Embedded in Every University of Surrey Degree

Surrey has undertaken a systematic redesign of every degree program to embed discipline-specific AI teaching, shifting assessment toward process over outputs. English literature students, for example, will submit annotated close-reading extracts alongside essays. The approach applies to current students, not just incoming ones — a level of institutional commitment few universities have executed.

💬 Editor’s note: The assessment shift — from evaluating what students produce to evaluating how they produce it — is definitely an important move in the right direction. It’s also probably very labor-intensive for both the faculty and students. I am wondering what faculty development and support infrastructures are in place to support such process-based assessments across all departments. In addition, how such assessements are designed to ensure students perceive them as meaningful. — TZ

3. FutureEd: 71 AI Education Bills Across 27 States

Legislative Tracker: 2026 State AI in Education Bills (FutureEd)

FutureEd’s 2026 tracker now monitors 71 bills across 27 states addressing AI in classroom instruction, up from 52 bills earlier in the session. Approaches range from AI literacy graduation requirements (Hawaii) to written parental opt-in consent (South Carolina) to district-level AI policies before 2027-28 (Oklahoma).

💬 Editor’s note: Whatever states decide for K-12 today shapes what universities receive tomorrow. — TZ

What’s in the News

1. Google Unveils Connected AI Tools for Classrooms at ISTE 2026

Building AI Tailored for Education, with Educators in the Lead (Google Blog)

Google announced a major expansion of AI tools across Google Classroom, Chromebooks, and Gemini at ISTE 2026 — a Classroom app in Gemini, “study notebooks” for personalized learning, and teacher-led AI activities grounded in school curricula. Google also announced funding for aiEDU to support Title I districts.

💬 Editor’s note: I wonder how these platforms and tools treat the student work submitted to them. On one hand, such data can help improve the tools; on the other, we need to address students’ governance and IP rights over their own data. See Student AI Bill of Rights. — TZ

2. Microsoft Releases Third Annual AI in Education Report at ISTE

Microsoft’s New AI in Education Report (Microsoft Source)

Alongside the survey data (covered in Research Highlights), Microsoft announced new AI features in Microsoft 365 Education at no additional cost: AI-assisted Unit Plans, Student AI Guidelines in Assignments, and a no-cost AI Literacy for Educators credential pathway.

💬 Editor’s note: The free credential pathway is worth watching. How would educational agencies and institutions recognize such a credential? — TZ

3. Times Higher Education: University AI Policies “Promise Support but Deliver Surveillance”

University AI Policies ‘Promise Support but Deliver Surveillance’ (THE)

THE’s coverage of the HEPI study (detailed in Research Highlights) emphasized the gap between policy rhetoric and policy function. Faculty shared it widely, many adding “this is us” commentary. University communications teams largely stayed silent — the finding is difficult to rebut without substantively redesigning policies.

4. AI Campus Index: First National Ranking of Universities on AI Readiness

AI Campus Index

The AI Campus Index is the first national platform ranking colleges and universities on how well they use and provide AI — across student access, teaching, research, governance, and operations. Rankings are based on publicly available data and self-reported institutional surveys. The index provides a comparative benchmark at a time when most institutions are making AI investments without clear metrics for progress.

💬 Editor’s note: A ranking inevitably shapes behavior. Worth watching how institutions respond. — TZ

Most Discussed

1. GPQA Diamond Saturation: AI Models Now 30 Points Above Human PhD Experts

GPQA Diamond Benchmark Leaderboard (Artificial Analysis)

Seven frontier AI models now score between 93.2% and 94.6% on GPQA Diamond — a graduate-level science benchmark where human PhD experts average approximately 65%. The benchmark is widely considered saturated, with top-model differences falling within measurement noise.

💬 Editor’s note: If exam performance can be replicated by a token-processing system, what is the exam actually testing? — TZ

2. Anthropic’s 2026 Agentic Coding Trends Report

2026 Agentic Coding Trends Report (Anthropic)

Anthropic’s report documents the shift from AI as coding assistant to autonomous agent team, arguing 2026 marks the transition where engineers move from writing code to orchestrating systems that write it. Developers use AI in ~60% of work but can “fully delegate” only 0-20% of tasks. Non-technical roles — product managers, designers, marketers — are also adopting coding agents.

💬 Editor’s note: I’ve started using agentic AI systems for what I call “natural language programming,” and it’s remarkably empowering. At the same time, I recognize how hard it can be to detect AI failure modes when the underlying software engineering details stay opaque. It’s difficult, for instance, to notice when an agent quietly takes a computational shortcut without alerting you, or when it drifts from your instructions because of context window limits. The rise of non-technical use of coding agents represents both exciting progress and a real need for training. — TZ

3. Brown Professor Suspects Majority of Class Used AI to Cheat on Take-Home Midterm

Brown Professor Suspects Most of His Class Used AI to Cheat (Inside Higher Ed)

Brown economics professor Roberto Serrano gave his first take-home midterm in nearly two decades — prompted by student anxiety after a campus shooting. The class had grown from ~30 to 86 students. The midterm average was 96%, far above the historical 65–80% range. After switching the final to in-person, the average dropped to 48.6% — a historic low. Eighteen students dropped the course, nine skipped the final, and 19 failed. Brown’s own generative AI committee, reporting the same week, found 75% of faculty concerned about AI cheating and recommended that faculty “de-emphasize punishment” while establishing clearer norms.

💬 Editor’s note: This case is an opportunity for an important conversation with our students. When we use AI, it becomes genuinely difficult to assess how much of the work we “own” — how much of the thinking, the problem-solving, the learning actually happened inside our own heads. Students aren’t necessarily trying to cheat; many may not fully realize how much cognitive work they’re offloading. What we can do is offer more low-stakes, self-assessment opportunities — practice problems, ungraded quizzes, reflective exercises — and remind students that the point of doing them without AI is to learn how much they’re actually learning. The gap between Serrano’s midterm and final scores is the gap between AI-assisted performance and demonstrated competence. Students deserve to see that gap for themselves, in low-stakes settings, before it shows up on an exam. — TZ

4. Early-Career Employment Decline in AI-Exposed Occupations

Software Developer Employment for Ages 22-25 Falls Nearly 20% Since 2022

Employment data shows that early-career workers in AI-exposed occupations — software development, clerical work, content creation — have experienced 16% relative employment declines since 2022, while employment for experienced workers remains stable. NBER projects approximately 502,000 AI-related job cuts in 2026, roughly 9x the estimated 55,000 in 2025.

Interesting Ideas & Repos

1. DeepTutor: Agent-Native Personalized Tutoring Platform

DeepTutor (GitHub)

An open-source, agent-native learning workspace from HKU connecting tutoring, problem-solving, quiz generation, and research. Features EduHub, a community hub for sharing teaching-oriented agent skills — Socratic tutors, flashcard builders, essay feedback, exam blueprints.

2. AI Engineering from Scratch: 503 Lessons, 320 Hours

AI Engineering from Scratch (GitHub)

A comprehensive open-source curriculum covering 20 phases from fundamentals to advanced AI engineering, with 503 lessons across ~320 hours. The repository has attracted 55,593 monthly visitors and 7.5K stars, suggesting significant community adoption.

3. LearnHouse: Open-Source Learning Platform with AI Integration

LearnHouse (GitHub)

A next-generation open-source learning platform featuring a block-based content editor, AI-generated interactive elements, code execution with auto-grading in 30+ languages, collaborative whiteboards, and context-aware AI for learning and teaching. An open-source alternative to proprietary LMS platforms.

4. Plano: AI-Native Proxy for Cost Control in Agentic Apps

Plano (GitHub)

An open-source AI-native proxy and data plane for agentic applications — with built-in orchestration, smart LLM routing, observability, and guardrail filters. Useful for institutions experimenting with multi-agent workflows who need cost control and model management without building custom infrastructure. 6.7K stars.

5. Speechmatics Academy: Open-Source Voice AI Examples and Tutorials

Speechmatics Academy (GitHub)

A comprehensive collection of working examples for speech-to-text and text-to-speech applications — from basic transcription to real-time voice agents, healthcare dictation, and multilingual support. Useful for faculty exploring voice-based AI tutors or accessibility tools. Includes integrations with LiveKit, Pipecat, and Twilio for building conversational AI applications.

Try This Week

AI Policy Audit: Is Your Syllabus Statement Educating or Surveilling?

Inspired by the HEPI study, try a simplified version of its analysis on your own AI syllabus statement. Time: 15-20 minutes.

Pull up your current AI syllabus statement (or department policy). If you don’t have one, that’s data too.
Count how many sentences describe: (a) what students should learn about AI, (b) how students are allowed to use AI, (c) consequences for misuse.
Calculate the ratio. The HEPI study found nearly half of educational-sounding policies were actually detection-and-discipline frameworks.
Draft one sentence describing an AI learning outcome for your course — something students should be able to do with AI by the end of the term.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

aiX Weekly — AI in Higher Education (July 8th, 2026)

Wed, 08 Jul 2026 00:00:00 +0000

This post tracks converging empirical evidence on AI and learning outcomes, SUNY’s systemwide AI policy across all 64 campuses, and the fast-growing market for detection-evasion tools.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

Reviewed by Tian Zheng on July 7th, 2026.

TL;DR

Two large-scale studies document a gap between homework scores and exam performance in AI-exposed courses — a 26,811-student longitudinal study and a Berkeley analysis of 500,000 grades both find the same pattern (Research Highlights)
SUNY adopts a systemwide AI policy across all 64 campuses — AI literacy required for all incoming undergrads starting Fall 2026 (Institutional Movements)
At least 150 AI “humanizer” tools now exist to evade detection software — drawing 33.9 million combined monthly visits (What’s in the News)
Students’ self-reported concern about AI’s impact on critical thinking rose 13 points in 10 months — from 54% to 67% (Most Discussed)

This Week at a Glance
Research Highlights
Institutional Movements
What’s in the News
Most Discussed
Interesting Ideas & Repos
Try This Week

This Week at a Glance

This week’s most notable development is converging empirical evidence on AI’s relationship to learning outcomes. Three large-scale studies — from ALEKS, Berkeley, and China — all find that AI-exposed courses show higher homework scores alongside lower exam performance. These are studies large enough to move beyond anecdote.

Relevant to faculty: The grade-and-learning research suggests that courses weighting unsupervised homework heavily may be measuring AI-assisted performance rather than student learning. Assessment design appears to be the key variable.

Relevant to institutional leaders: SUNY’s 64-campus policy and the Student AI Bill of Rights highlight the gap between AI adoption and governance infrastructure.

Relevant to students and researchers: The RAND finding that student concern about AI’s cognitive impact is rising alongside usage adds an important self-awareness dimension.

Research Highlights

1. Faster Completion, Less Learning: AI Reduces Study Time and Knowledge Retention

Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build

A ten-year panel analysis of 3.2 million ALEKS learning interactions found that after ChatGPT’s release, college students’ study time on AI-susceptible math problems declined 2.8% per quarter, cumulating to a 26.9% reduction over eleven quarters. Retention testing showed a 25% cumulative decline in odds of correct response — and this divergence vanished under proctored conditions, ruling out genuine efficiency gains. The age gradient is informative: fifth graders, least likely to use AI independently, showed no detectable effect.

Longitudinal panel analysis

💬 Editor’s note: I found The proctored vs. unproctored finding to be interesting and well-studied to distinguish genuine efficiency from substitution. For curricular design, this is useful evidence for thinking about where unsupervised practice fits in a course. This study also offers a great discussion case study on observational study, experimental design, hypothesis testing at all levels of the statistics curriculum. — TZ

2. AI Grade Inflation Documented Across 500,000 Grades

Artificial Intelligence and Grade Inflation (UC Berkeley CSHE Working Paper)

Analyzing over 500,000 grades from 2018–2025 at a large Texas research university, Berkeley researchers found that AI-exposed courses saw A grades rise by 13 percentage points — roughly 30% above the 2022 baseline. The increases concentrated in writing and coding courses, and were larger where homework carried greater weight. The homework-weight finding identifies a concrete design lever faculty can adjust.

Observational study with difference-in-differences

💬 Editor’s note: I like how the author used a LLM to process all syllabi when determining a course’ AI exposure as a way to scale the study. The concentration of grade inflation in writing and coding — the domains where AI tools are most capable — is informative for curricular planning. The homework-weight mechanism is a particularly useful finding: it identifies something faculty can act on directly. The Chronicle covered this study and it’s generated substantive discussion. — TZ

3. Chinese Study of 26,811 Students: AI Cuts Homework Time, Tanks Exam Performance

Study Links AI-Assisted Homework to Lower Exam Scores (Secondary source — the underlying study is a CEPR discussion paper; link the primary source when available.)

Tracking 26,811 secondary students over 30 months, researchers found that generative AI reduced homework completion time by ~30% and increased homework scores by 18%, but monthly exam scores decreased by ~20% within six months. High-stakes entrance exam penalties reached 18–24% over two years. Roughly 80% of learning losses stemmed from a “fast completion + high score” behavioral signature — a practical detection pattern that doesn’t rely on AI detection tools.

Longitudinal cohort study

💬 Editor’s note: Three studies in this post all point in a similar direction: a measurable gap between performance signals and assessed competence. The “fast completion + high score” behavioral marker offers a practical insight for curriculum design. Turning knowledge exposure into skills and intuition require slowness and frictions beyond conventional homework and exams. — TZ

Institutional Movements

1. SUNY Adopts Systemwide AI Policy Across All 64 Campuses

SUNY Sets Systemwide AI Policy

The State University of New York Board of Trustees adopted a systemwide AI policy requiring all 64 campuses to adopt or update AI guidelines by December 31, 2026. AI literacy becomes part of general education for all incoming undergraduates starting Fall 2026. The policy mandates bias evaluation for AI tools and strengthened data privacy protections.

💬 Editor’s note: SUNY’s approach through 20-member AI for Public Good Fellows cohort is a promising peer-led implementation mechanism for an institutional response. Phil Hill at One EdTech noted offered a critical analysis that raised important questions, which is worth reading about. — TZ

2. Entry-Level Job Market Shifts and Higher Education

How AI Broke the Entry-Level Job (Washington Monthly)

The economy has added 3 million white-collar jobs since ChatGPT arrived, yet new graduates face a tightening on-ramp. The Washington Monthly describes an “experience creep” pattern: employers demanding proven experience for roles once open to new graduates, as AI handles the tasks that used to be their training ground.

💬 Editor’s note: This story connects the learning gap research to workforce trends. I have shared at a few discussion and panels that I believe employers have need for fresh perspectives and talents. However, just like AI disrupted teaching and learning, it has also disrupted talent recruitment and evaluation. As disciplines continue to re-focus their expertise and knowledge mission with AI being part of their tool kit and workflows, faculty will design courses around real AI native projects with real stakes. This may potentially address the entry-level gap from the curriculum side. — TZ

3. Student AI Bill of Rights Unveiled

Consumer Protection Group Unveils Student AI Bill of Rights

The National Student Legal Defense Network released a “Student AI Bill of Rights” establishing five articles: transparency (know when AI evaluates you), human oversight and appeal, data sovereignty and intellectual property, freedom from algorithmic bias, and AI-informed education. The framework is designed to guide institutional policy development.

💬 Editor’s note: Principles on the protection of students should be central to our adoption of AI in teaching and learning. Article III — the right to data sovereignty and IP — is in particular very important. It asserts that enrollment does not constitute consent to commercialization of student work. As institutions sign deals with AI companies, the question of what happens to student data and coursework is becoming more and more relevant. We need more technical understanding of the data (raw and processed) flow, compute and storage, and memory retention inside these tools. — TZ

What’s in the News

1. Senate Holds First Hearing on AI in K-12 Classrooms

‘AI is here’: Lawmakers pressed to prepare students for future that’s already arrived (Washington Times)

The Senate HELP Subcommittee held a June 16 hearing on AI in K-12. Witnesses urged both guardrails and investment. FutureEd has been tracking AI-in-education bills across states; recent counts suggest over 60 bills in more than 25 states.

💬 Editor’s note: K-12 hearings matter for higher ed. The students arriving on campus in 2028 will have been educated under whatever AI policies these state bills produce. What happens in state legislatures this year could shape what happens in our classrooms in two. — TZ

2. NPR/Ipsos Poll: Teachers Say AI’s Impact Will Eclipse the Internet

Poll: Teachers worry AI is impacting students’ critical thinking (NPR/Ipsos) (Primary survey source: Ipsos)

A nationally representative poll of 545 K-12 teachers found that nearly three-quarters believe AI’s impact on education will be more significant than the internet. Over half (54%) say AI makes it harder for students to learn critical thinking. Nearly 60% say AI is eroding trust between students and teachers. Only about a third say their school has formal AI guidelines.

💬 Editor’s note: I found the trust erosion result most noteworthy. Violations of academic integrity long predate AI. The widespread tension between cheating with AI and the return of paper-based tests for evaluation is not just a strain on the relationship between students and their teachers. It is rooted in a confusion, shared by educators and students alike, about the content and values of teaching and learning in the face of AI’s capabilities. Would students cheat with AI if they were doing exercises for their own test prep? — TZ

3. NBC News: Students Use AI “Humanizer” Tools to Beat Detection

To avoid accusations of AI cheating, college students are turning to AI (NBC News)

At least 150 AI “humanizer” tools now exist to rewrite AI-generated text to evade detection software. NBC reports that 43 such platforms drew 33.9 million combined website visits in October 2025, with some charging $20–50/month. The tools manipulate the statistical signals detection software analyzes.

💬 Editor’s note: I believe most faculty now realize that AI detection does not work as intended. — TZ

4. The Economist: AI Models’ Values Don’t Reflect Most of the World’s

AI models’ values are very different from most people’s (The Economist)

The Economist tested 25 frontier AI models against the World Values Survey — the same instrument used since 1981 to map cultural values across 100+ countries. The models overwhelmingly cluster in the quadrant populated by rich Western nations. OpenAI’s GPT models score more secular than any country on earth; Google’s Gemini models emphasize individual freedom more than any human population surveyed. No model reflects the worldviews of most African or Muslim countries. Chinese models like DeepSeek show distinctly different value alignments.

💬 Editor’s note: An AI tutor trained to emphasize individual self-expression may give quite different guidance than one oriented toward collective responsibility — and neither is “wrong.” The pedagogical question is whether students understand that AI advice comes with embedded values. — TZ

Most Discussed

1. Reddit Study: 270K Posts Map the Evolution of AI Education Discourse

ChatGPT vs Teachers vs Students: Large-Scale Analysis of Generative AI Discourse in Education Communities on Reddit

Researchers analyzed 270,000 AI-related Reddit posts from 26 education subreddits spanning November 2022 to April 2026. Misconduct Enforcement comprised 12.1% of discourse, AI Detection/False Accusations 10.8%, and Assessment Redesign 5.5%. The study documents a clear evolution: from an early detection-and-evasion arms race to sustained enforcement, with constructive integration only beginning to challenge that framing in mid-2024.

💬 Editor’s note: According to the study, the shift from academic integrity concerns to constructive integration took 18 months after ChatGPT’s release. This was not simply due to a slow reaction by higher education. AI’s rapid development kept moving the target: it went from being a convenience tool to a disruptive force in knowledge generation and professional work. In some domains, AI tools are now mature and well integrated into practice; in others, they remain unreliable. Many of these cases require active research before we can conclusively say what should change in our curriculum. The real challenge is how to react more efficiently to a landscape that keeps shifting. — TZ

2. August AI Scores 100% on USMLE

August Benchmark 2026: 100% on USMLE

August AI reports achieving a perfect score on the U.S. Medical Licensing Exam, claiming to outperform GPT-5 and other frontier models. Coming on top of AI’s near-ceiling performance on bar exams, CPA exams, and MMLU-Pro, the result reignited debate about professional credentialing when AI can pass every exam.

💬 Editor’s note: The common thread across these benchmarks: assessments that primarily measure knowledge recall and procedural application are now within AI’s capability range. The dimensions that remain distinctly human — judgment under uncertainty, ethical reasoning in context, creative synthesis — may deserve more weight in both education reform and assessment design. — TZ

3. Students’ Self-Awareness About AI Dependency

RAND findings on student concern about critical thinking | EdWeek: Students Are Worried That AI Will Hurt Their Critical Thinking Skills

The RAND American Youth Panel found that student concern about AI harming critical thinking rose from 54% to 67% in just 10 months, even as AI homework use rose from 48% to 62%. The 13-point jump alongside rising usage suggests students are reflecting on trade-offs even as they continue using the tools.

💬 Editor’s note: Faculty and students share the same concern! Integrating AI into teaching and learning and set up AI policies as intentional learning design — “this assignment limits AI use because the cognitive work is the point” — may find a more receptive audience than they expect. — TZ

Interesting Ideas & Repos

1. FabData-LLM: RAG System for Educational Document Collections

AI-for-Education/FabData-LLM-retrieval on GitHub

A full end-to-end platform for building a Retrieval-Augmented Generation system from a catalog of documents. Faculty could build a course-specific AI tutor that only draws from assigned readings, ensuring students engage with intended material while still using AI as a study tool.

2. Socratic AI Design as Pedagogical Template

Claude’s Learning Mode: Transform AI into a Socratic Tutor

Anthropic’s Learning Mode — which replaces direct answers with Socratic questioning — is increasingly cited not as a product feature but as a design template. Even faculty not using Claude can adopt the principle: any AI interaction can be structured to ask questions rather than provide answers, using Socratic system prompts.

Try This Week

Check your course’s “AI vulnerability ratio.” Pick one course you’re teaching or planning for fall. List every graded assignment and classify each as: (1) supervised — completed in class or under observation; (2) unsupervised + AI-resistant — requiring something AI can’t provide (personal reflection, fieldwork observation, original data); (3) unsupervised + AI-susceptible — feasibly completable by AI.

Calculate what percentage of the final grade comes from category 3. The Berkeley study found grade inflation concentrated where homework carried greater weight; the Chinese study found exam scores dropped 20% when students relied on AI for unsupervised work.

If a substantial portion of your grade comes from category 3, options include adjusting its weight, adding verification mechanisms (brief oral defense, in-class follow-up), or redesigning assignments to fall into category 2.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

aiX Weekly — AI in Higher Education (July 1st, 2026)

Wed, 01 Jul 2026 00:00:00 +0000

Welcome to the inaugural issue of aiX Weekly — a curated digest of research, institutional moves, and debate at the intersection of AI and higher education.

Each issue is organized around a set of recurring sections and pairs with our companion AI and Higher Education timeline, which traces the broader arc of how AI has reshaped higher education since late 2022.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

Reviewed by Tian Zheng on July 1st, 2026.

TL;DR — Questions This Issue Helps You Think About

How do we measure AI competency among faculty? → New FALCON-AI scale attempts to offer an instrument (Research Highlights)
What does the evidence say about AI and critical thinking? → New studies find a correlation with cognitive offloading — but self-efficacy may be a moderating factor (Research Highlights, Most Discussed)
How are large-scale AI deployments playing out? → Cal State’s $17M OpenAI contract is generating discussion about governance and consultation (What’s in the News)
What’s the conversation around faculty autonomy in AI adoption? → Writing teachers passed a resolution on the right to opt out; the broader discussion continues (What’s in the News)
What are we learning about AI detection tools? → Accuracy and bias concerns are prompting scrutiny, lawsuits, and calls for review (What’s in the News)
What does a campus-wide AI fluency requirement actually look like? → Ohio State’s roadmaps across all colleges provide a model (Institutional Movements)
How are other countries approaching this? → China just mandated AI as a core university course nationwide (Institutional Movements)
What tools are graduate students exploring for research? → Tools gaining traction include Elicit, Consensus, and NotebookLM (Interesting Ideas & Repos)

This Week at a Glance
Research Highlights
Institutional Movements
What’s in the News
Most Discussed
Interesting Ideas & Repos
What Changed
Try This Week

This Week at a Glance

This week’s stories illustrate a field navigating the distance between AI adoption and the frameworks meant to support it.

New research is adding definition to the picture: cognitive offloading from AI use correlates with lower critical thinking scores, but a CMU/Microsoft finding — that self-confidence (not tool-confidence) is associated with more critical thinking — suggests pedagogical design may matter as much as the tools themselves. The FALCON-AI scale and the “AI Literacy Heptagon” offer new instruments for measuring AI competency among faculty and students.

On the ground, institutions are working through familiar tensions. Cal State’s $17M OpenAI contract drew pushback from faculty and students. Writing teachers voted on a resolution supporting the right to opt out. AI detection tools face growing scrutiny over accuracy and bias. And 90% of faculty express concern about AI’s impact on critical thinking, even as institutions from Stanford to Ohio State move forward with AI integration.

Relevant to faculty: Research increasingly suggests that how AI is integrated matters more than whether it is. Building student self-efficacy — not just tool fluency — appears linked to preserving critical thinking. The CCCC opt-out resolution reflects ongoing discussion about disciplinary autonomy in AI adoption.

Relevant to institutional leaders: The Cal State and NYC experiences highlight questions about sequencing — when governance frameworks and faculty consultation happen relative to deployment. The IREX finding that only 1 in 3 universities has a clear AI strategy offers a useful benchmark.

Relevant to students and researchers: Several AI research tools are gaining traction among graduate students (Elicit, Consensus, NotebookLM). Meanwhile, accuracy concerns around AI detectors — especially regarding non-native English speakers — are worth following.

Research Highlights

FALCON-AI: A New Scale for Measuring Faculty AI Competency

Instrument Development · Empirical Evidence

Institutions are pouring resources into faculty AI training, but how do you know it’s working? Existing AI literacy instruments were built for students or the general public and miss the role-specific demands faculty face — teaching with AI, researching with AI, and governing its use in their departments are fundamentally different tasks.

Song, Moon, Yang & Kilgore (2026) developed the FALCON-AI Scale, a psychometrically validated instrument designed specifically for university faculty — addressing a gap the authors identify in existing instruments, which lacked role-embedded faculty indicators. Grounded in the Critical Tech-resilient Literacies (CTRL) framework, it maps 43 items across three literacies (functional, evaluative, ethical) and four faculty work domains (general, teaching, research, service).

Source: arXiv:2603.20220

💬 Editor’s note: As a statistician, I find this kind of instrument development genuinely encouraging — we need validated measures before we can rigorously evaluate whether our faculty development efforts are working. What needs to happen next is meaningful predictive validity: do FALCON-AI scores actually correlate with observable differences in how effectively faculty integrate AI? Self-report is a good starting point, and pairing it with behavioral or outcome measures would strengthen the case. For measuring the usefulness of AI related faculty development programs, this could be a useful pre/post tool worth piloting. — TZ

AI-Overdependence and Cognitive Decline: New Evidence

Systematic Review + Empirical Study

The central anxiety in AI education is whether AI tools are eroding the cognitive capacities they’re supposed to support. Faculty sense it anecdotally — students seem less willing to struggle with hard problems — but the field has lacked synthesized evidence.

A 2026 review in Computers in Human Behavior Reports synthesizes findings suggesting that higher AI use correlates with greater cognitive offloading and lower critical thinking scores, with younger users most affected. A complementary Carnegie Mellon/Microsoft Research study reveals a crucial paradox: confidence in GenAI tools was associated with less critical thinking, while self-confidence was associated with more.

Source: Computers in Human Behavior Reports | Center for Engaged Learning

💬 Editor’s note: The correlation between AI use and reduced critical thinking is a finding worth taking seriously, but as a statistician I want to emphasize the causal caveat here — correlation studies cannot tell us the direction. It’s entirely plausible that students with weaker critical thinking skills are more likely to rely on AI, rather than AI causing the decline. We need randomized or longitudinal designs to disentangle this. That said, the self-confidence finding is intriguing and practically useful. In my own teaching, I’ve found that building students’ confidence through structured hands-on problem-solving changes how they engage with any tool, AI included. We need to design AI systems and AI learning experiences that can help develop cognitive skills rather than replacing or supressing the use of these skills. — TZ

Critical Inker: Scaffolding Critical Thinking in AI-Assisted Writing

System Design · Preprint · Interesting Concept

If students are going to use AI for writing regardless of policy, the question becomes: can we design the interaction itself to preserve cognitive engagement? Most AI writing tools optimize for output quality, not learning — the student gets a better essay but does less thinking.

Critical Inker takes a different approach: it uses Socratic questioning to interrupt the passive consumption of AI-generated text, prompting students to evaluate, question, and refine. The system intervenes at the point of cognitive offloading rather than trying to prevent AI use altogether.

Source: arXiv:2604.07167

💬 Editor’s note: I like the design thinking here — building reflection into the AI interaction rather than policing AI use after the fact. Reflection and Socratic questioning are both recurring design choices among the aiX program projects. The learning happens in the friction, not in the output. — TZ

AI Competency Strategies via LLM-Based Delphi Method

Exploratory / Methodological

Identifying what AI competencies matter in higher education usually requires expensive, slow expert consensus processes. Can AI accelerate the process of defining what humans need to know about AI?

This Frontiers in Education study replaces human expert panels with an LLM-based Delphi methodology to identify essential AI competencies, examine integration barriers, and propose strategies.

Source: Frontiers in Education

💬 Editor’s note: This is a creative methodological experiment, and I appreciate the attempt to speed up consensus-building. The meta-question — using AI to study AI education — is itself worth discussing with faculty as a case study in how to leverage AI rigorously and effectively in higher education. — TZ

Capability-Based Training Framework for GenAI in Higher Ed

Conceptual Framework / Literature Review

*Most AI literacy frameworks stop at “understanding” — can you define what a neural network is, can you identify bias. But faculty and students don’t just need to understand AI; they need to use it effectively within their disciplines. The gap between knowing what AI is and knowing how to apply it in your field is where most training programs fall short.

This Frontiers in Education framework proposes moving from literacy to capability — structured training that develops the ability to use generative AI in discipline-specific contexts, not just comprehend it abstractly.

Source: Frontiers in Education

💬 Editor’s note: The literacy-to-capability shift resonates with how I envisioned the aiX program. The shift suggests training should be co-designed with disciplinary faculty, not delivered generically. In Statistics, knowing what a p-value is differs enormously from knowing when and how to use one in one’s research. The same applies to AI. Conceptual frameworks like this are useful for orienting program design. Our aiX program helps us explore what does “capability-based AI training” look like in a humanity department vs. a professional school. The cross-disciplinary community of the aiX program adds values to such frameworks. — TZ

Institutional Movements

Stanford: $1M Seed Grants for AI + Education Research

The need: Most AI-in-education evidence is either too technical (model performance) or too anecdotal (observational data via survey). There’s a gap in rigorous, pedagogy-centered research on how AI actually changes learning.

Stanford’s AIMES initiative and the Accelerator for Learning announced $1 million in seed grants for course development and research. A notable design choice: the call explicitly welcomes proposals from faculty skeptical of AI, not just advocates.

Source: Stanford Report

💬 Editor’s note: The “skeptics welcome” framing is worth highlighting. Inviting skeptical faculty to study AI’s effects — not just implement AI tools — is how we build a research base that includes null results and failures, which are just as important as successes. — TZ

Columbia: Reimagining Teaching and Learning Forum

The need: Faculty experimenting with AI in their courses often work in isolation — there’s no natural venue to see what colleagues across disciplines are trying, what’s working, and what’s failing.

Columbia’s CTL hosted “Reimagining Teaching and Learning in the Age of AI,” built around a Demo Expo where faculty showcased live AI-enabled course projects. Stakeholders ranged from educational leadership to students and external experts.

Source: Columbia CTL

💬 Editor’s note: Events like this are where the most valuable learning happens — faculty seeing each other’s experiments, not just hearing about AI in the abstract. I attended this forum and came away with several ideas I’m exploring in my own courses. aiX program organizes cohort-based collaborations — where faculty from different departments co-develop and evaluate AI-integrated assignments over a semester, a combination of faculty development and evidence-building. — TZ

Ohio State: AI Fluency Initiative Reaches Milestone

The need: Requiring “AI fluency” is easy to announce; defining what it means in 200+ majors across 15 colleges is the hard part.

Ohio State pushed roadmap development to the college level: all colleges have now produced academic roadmaps showing how undergraduates in every major will build AI skills. The three-pillar model — foundational understanding, disciplinary application, ethical/societal impact — provides a common structure while allowing disciplinary customization.

Source: Ohio State Academic Affairs

💬 Editor’s note: The decentralized approach here — common framework, discipline-specific execution — mirrors how I have used to think about integrating data science literacy or quantitative reasoning across disciplines (at LEAP, DSI, through the collaboratory program and the aiX program, etc). — TZ

AAC&U: Institute on AI, Pedagogy, and the Curriculum (2026–27)

The need: Individual institutions redesigning pedagogy for AI in isolation risk reinventing the same wheels — rethinking assessment, navigating academic integrity, training faculty. A cross-institutional learning structure accelerates the work.

AAC&U’s seven-month, team-based Institute reports having served 316 teams from 296 institutions over two years. Stakeholders: campus teams typically include faculty, administrators, and instructional designers.

Source: AAC&U

💬 Editor’s note: 296 institutions working through the same set of problems together is an extraordinary resource for cross-institutional learning. If even a fraction of participating teams published case studies with outcome data, the field would have a much richer evidence base for AI pedagogy decisions. — TZ

China: National “AI + Education” Action Plan Through 2030

The need: Tech sector is advancing AI capabilities faster than the education system can produce people to build, govern, and critically evaluate them.

China’s Ministry of Education’s April 2026 action plan mandates AI as a basic public course in all universities, requires new interdisciplinary AI majors, and embeds AI in teacher qualification exams. The plan integrates research universities, tech enterprises, and national labs. Stakeholders: MoE, university leadership, and industry partners in a state-coordinated model.

Source: S&T Daily | Global Times

💬 Editor’s note: The scale of this initiative is remarkable and worth watching closely. — TZ

IREX/Development Gateway: Global AI Readiness Survey

The need: Universities are adopting AI tools rapidly, but “adoption” and “readiness” are not the same thing. Institutions needed a benchmark to understand where they actually stand relative to peers.

This global survey (Nov 2025–Jan 2026) found that only 1 in 3 universities has a clear AI strategy, fewer than 1 in 5 have governance structures, and only 37% of respondents reported receiving ongoing AI-related professional development. Stakeholders: university leadership, faculty, and IT administrators across multiple countries.

Source: IREX

💬 Editor’s note: There are both a gap and an opportunity — the institutions that invest in sustained, discipline-specific AI training for faculty are likely to see meaningfully different outcomes. — TZ

What’s in the News

NPR: Cal State’s $17M OpenAI Deal Draws Faculty and Student Discussion

California State University — the largest public university system in the U.S. — signed a $17 million contract with OpenAI for ChatGPT Edu, then renewed for another $13M/year over three years. NPR reports that the rollout generated significant discussion among faculty and students. A survey of over 94,000 people found that majorities of both groups expressed skepticism about AI’s educational benefits. The story has become a reference point in discussions about the sequencing of institutional AI decisions.

Source: NPR | VPM

Reception: Widely shared across faculty networks. Some commentators drew parallels to previous large-scale ed-tech rollouts; others defended the move as necessary experimentation at scale. The discussion highlighted different views on whether AI adoption is primarily an infrastructure decision or a pedagogical one.

What’s interesting here: The story surfaces a governance question many institutions are working through — how and when faculty and students are consulted in large-scale AI decisions. Different institutions are finding different answers.

💬 Editor’s note: The conversation about governance is important; equally important is designing these deployments so we actually learn something from them. — TZ

Inside Higher Ed: Writing Faculty Push for the Right to Refuse AI

The Composition and Communication Teachers of America (CCCC) overwhelmingly approved a resolution at its annual convention supporting faculty’s right to opt out of using generative AI in the classroom. The resolution reflects deep concern among writing educators that AI undermines the cognitive process of drafting, revising, and thinking through prose — the very skills their courses are designed to develop.

Source: Inside Higher Ed

Reception: The resolution prompted discussion across faculty networks. Supporters emphasized that writing pedagogy centers on the process of thinking, not the product. Others noted the tension with preparing students for AI-integrated workplaces. The conversation has become part of a broader discussion about disciplinary autonomy in AI adoption.

What’s interesting here: The debate reflects different views of what writing courses are designed to develop — cognitive process vs. communication product. Both perspectives have merit, and the field is still developing shared language for navigating between them.

💬 Editor’s note: I respect the reasoning behind this resolution. In statistics, the process of working through a problem — trying approaches, hitting dead ends, debugging your logic — is where learning happens. The final answer is almost beside the point. I can see why writing faculty feel the same way about drafting. The challenge is that this framing applies differently across disciplines: in some fields, using AI to handle routine production so students can focus on higher-order analysis may actually serve learning goals. What we need are discipline-specific conversations about where the cognitive work lives in each curriculum. — TZ

Chronicle & Inside Higher Ed: Faculty Are Overwhelmed — and Not in a Good Way

A January 2026 national survey found that 78% of faculty say AI-driven cheating is on the rise, but they are deeply split on what counts as cheating — just over half said following a detailed AI-generated outline is cheating; just under half said it’s legitimate or they’re unsure. Meanwhile, 95% say AI will increase students’ overreliance on tools over time. The data reflects a profession navigating significant uncertainty about definitions and expectations.

Source: Inside Higher Ed | Chronicle of Higher Education

What’s interesting here: The split on “what counts as cheating” may be the most telling finding. Shared definitions of legitimate AI use are still emerging — not just across institutions, but within departments. Faculty are making individual judgment calls, and students are navigating varied expectations. Developing shared norms, beyond formal policies, remains an open challenge.

💬 Editor’s note: The 50/50 split on whether an AI-generated outline counts as cheating is striking. If we don’t resolve this among faculty instructors, it means students in the same department may face opposite rules depending on which section they enrolled in. In my own thinking about AI policy for my introduction to statistics course, the most productive way starts with the intended learning outcomes of an assignment, here’s what a student could do with AI, which uses replace learning and which ones don’t? — TZ

False Accusations: AI Detectors and the Students Caught in Between

Multiple evaluations have found that AI detection tools struggle with accuracy, particularly on paraphrased AI content, and a Stanford study found they disproportionately flag non-native English speakers, whose structured prose can resemble AI-generated text. Turnitin claims a false positive rate below 1%, but acknowledges the tool should not be sole evidence in academic integrity cases. A Palo Alto family filed a civil rights suit after a false accusation; a non-native English speaker sued Yale alleging discriminatory treatment. Institutions have varied widely in how they deploy these tools — thresholds, review processes, and appeals paths differ significantly. UK universities saw AI misconduct cases rise nearly 400% in three years, though the role of detection tool accuracy in that increase is debated.

Source: SF Standard | Stanford EE | Strauss Troy

Reception: The topic has generated sustained discussion across faculty networks and social media. Some educators have called for pausing AI detection tools until accuracy and bias are better understood. Students have shared personal accounts of false accusations. Legal scholars have noted the evolving liability landscape.

What’s interesting here: The situation illustrates broader questions about deploying automated tools for high-stakes decisions before validation is complete. The disproportionate impact on non-native English speakers has drawn particular attention. Both research findings and legal proceedings are contributing to how institutions are re-evaluating their approaches.

💬 Editor’s note: As a statistician, the detection accuracy numbers are the heart of this story. Tools with documented accuracy limitations and bias against non-native speakers raise serious questions about their use in high-stakes decisions about students’ academic standing. This is a classification problem for high-stake decisions with well-understood tradeoffs between false positives and false negatives, which require ethical principles and debates. — TZ

NYC Schools: Parents Demand AI Pause Ahead of Governance Framework

New York City’s Department of Education has been developing AI guidance for its school system — allowing teachers to use AI for brainstorming and lesson planning, while restricting AI for grading, disciplinary decisions, or biometric data collection. Reports indicate that parents have raised concerns about AI deployment timing, calling for stronger governance frameworks before broader rollout.

Source: chalkbeat

What’s interesting here: Parents are emerging as an active stakeholder group in AI governance discussions. In higher ed, a parallel may be developing as students and families form views on how AI is used in instruction. The “pause until governance is ready” position reflects a sequencing question many institutions are navigating.

💬 Editor’s note: Governance first or adoption first? There’s not a single right answer. AI is developing quickly. There is need for piloting AI tools in contained settings while governance frameworks are being developed. Either full deployment or full pause seems right. — TZ

Most Discussed

Stanford HAI 2026 AI Index: Education Chapter

The report finds that four out of five U.S. high school and college students use AI for schoolwork, and that master’s graduates in AI-related fields rose 17% from 2023–2024. The report also examines how students are using AI, including for content creation and analysis — activities traditionally used to measure learning.

Why it caught attention: The scale. This is among the most comprehensive datasets on AI adoption in education, and the “four out of five” figure shifts the conversation from whether students use AI to how they use it. Who’s paying attention: administrators referencing it for strategic planning, faculty citing it in assessment redesign discussions, and ed-tech companies noting market growth. What it suggests: adoption has moved quickly, and the question of whether institutions can redesign learning to match — so that students creating content with AI are still developing the skills that content creation was designed to teach — is becoming central.

Source: Stanford HAI AI Index 2026

💬 Editor’s note: The “four out of five” adoption number is interesting. However, as AI gets more uses in learning, it is important to further differenting use cases. “Using AI for schoolwork” covers everything from asking ChatGPT to explain a concept (probably fine) to having it write an essay (probably not fine, depending on the assignment). The question isn’t whether to allow AI; it’s whether our assessments still measure what we think they measure for the learners. — TZ

The 90% Faculty Concern

National survey data showing 90% of faculty believe AI will decrease students’ critical thinking is driving vigorous debate. An MIT study finding that brain activity was suppressed during AI-assisted essay writing (without proper guidance) has become a widely cited data point.

Why it caught attention: The near-unanimity — 90% agreement among faculty on any topic is notable, and the number has become a reference point in discussions about AI’s cognitive impact. Who’s paying attention: faculty across disciplines, administrators, and researchers examining survey methodology. What it suggests: the MIT study adds nuance — brain activity was suppressed without proper guidance, suggesting that pedagogical design may moderate the effect. The 90% figure captures a widespread concern; the accompanying research points toward possible design responses.

Source: UW Arts & Sciences | EdWeek

Reception: The 90% figure has been cited both as an important signal and questioned as a function of how the survey was framed. Some argue the real variable is assignment design, not AI itself. Others point to the MIT brain-activity data as evidence worth taking seriously. The conversation continues to evolve.

💬 Editor’s note: The MIT finding is where the actionable insight lives: brain activity was suppressed without proper guidance. It suggests the variable we can control is pedagogical design, not AI access. — TZ

OECD Digital Education Outlook 2026

The OECD report examines how GenAI tools interact with teaching expertise. Among its findings: GenAI may help amplify teachers’ capacity when integrated with their expertise, with suggestive evidence that less experienced tutors can benefit from AI support.

Why it caught attention: It offers a nuanced framing: AI may raise the floor for less experienced educators without diminishing the effectiveness of strong ones — though the specific evidence base for this claim warrants closer reading of the full report. Who’s paying attention: policymakers and institutional leaders interested in evidence beyond efficiency gains, and faculty development directors exploring how training and AI tools interact. What it suggests: the interaction between expertise and AI may be more important than AI alone. The finding is consistent with investing in pedagogical expertise first, then exploring how AI tools complement it.

Source: OECD

Interesting Ideas & Repos

AI-First Curriculum Design

Why it’s interesting: Most institutions are bolting AI onto existing courses as an afterthought. This Faculty Focus piece articulates what it looks like to design courses with AI as a foundational assumption — AI handles repetitive mechanics (grammar feedback, rubric alignment) while faculty focus on mentoring and critical dialogue. The shift from “AI-permitted” to “AI-first” is a meaningful design philosophy change worth watching as it develops.

Source: Faculty Focus

Microsoft: Generative AI for Beginners (GitHub)

Why it’s interesting: A 21-lesson open course covering prompt engineering through RAG pipelines, agents, and deployment. It’s beginner-friendly but technically substantive — the sweet spot for faculty who want to understand what their students are actually doing with AI tools, not just read about it. Could serve as the backbone of a faculty development workshop series, and the open license means you can adapt it.

Repository: microsoft/generative-ai-for-beginners

LLMs-from-Scratch

Why it’s interesting: Builds a GPT-style model from scratch in PyTorch. In a research training context, this is uniquely valuable — it replaces the “magic black box” understanding of AI with mechanical comprehension. Faculty and graduate students who work through this will be better equipped to critically evaluate AI outputs because they understand what the model is actually doing. Pairs well with the cognitive offloading research above: deeper understanding may build the self-confidence that protects against uncritical AI dependence.

Repository: rasbt/LLMs-from-scratch

Open-Source AI Tutor with Spaced Repetition

Why it’s interesting: The growing ecosystem of open-source AI tutors exploring spaced repetition, personalized examples, and adaptive pacing. What makes the space worth watching is the design pattern — learning tools that adjust to individual pace rather than delivering uniform content. For anyone designing AI-augmented learning experiences, browsing active projects in this space is more productive than starting from scratch.

Browse: github.com/topics/ai-tutor (Topic page — browse for specific projects of interest.)

PhD Research Tool Stacks for 2026

Why it’s interesting: Several AI research tools are gaining traction among graduate students, including Elicit (systematic literature search), Consensus AI (evidence-based answers from peer-reviewed research), and NotebookLM (synthesis and note-taking). This matters for research training because the tools students choose shape the research they produce — and most doctoral programs aren’t teaching tool selection as a methodological skill. The key caveat: AI accelerates literature discovery but cannot replace manual verification. Programs that teach the stack without teaching verification are training speed without rigor.

Source: Thesify (blog) (Note: vendor/blog source — verify specific tool claims against the tools’ own documentation.)

What Changed

This section tracks what’s notable this week in the context of the broader timeline. Future issues will note when earlier stories develop, when sources are corrected, or when our framing needs revisiting.

This week’s stories connect to several threads on the aiX Timeline:

The evidence base is growing. The cognitive offloading and FALCON-AI studies join a building body of research that started arriving in earnest in early 2025. Most findings remain correlational — the field is still waiting for experimental and longitudinal designs.
Governance questions are becoming concrete. Cal State’s contract discussion and the CCCC resolution move governance from abstract principle to specific institutional decisions. The IREX benchmark (1 in 3 with a clear strategy) provides a reference point.
Detection remains unsettled. The accuracy and bias concerns around AI detectors continue to develop, now with legal proceedings adding a new dimension to what began as a validation question in 2023.
International approaches are diverging. China’s nationwide mandate and the OECD’s expertise-interaction finding offer two very different data points for institutions considering their own directions.

Try This Week

Audit one assignment for AI vulnerability. Pick a single assignment from your current or upcoming course. Ask yourself: could a student complete this entirely — and competently — using ChatGPT or Claude? If the answer is yes, consider whether the assignment is measuring something AI can already do. Try redesigning it to include something AI can’t provide: a personal observation, a judgment call grounded in disciplinary experience, a connection to class discussion, or an original dataset. The goal isn’t to make the assignment “AI-proof” — it’s to make it worth doing even if AI exists.

Curated by Claude for the aiX Programs, Columbia University. AI can make mistakes and so is the human reviewer. Please double-check the linked sources.

AI and Higher Education: A Brief Timeline

Tue, 23 Jun 2026 00:00:00 +0000

An interactive timeline tracing pivotal moments in AI and higher education — from ChatGPT’s launch through exam disruptions, campus tensions, workforce shifts, and the emergence of institutional AI strategies.

This timeline is a companion resource to the aiX Weekly Digest, curated to help anyone who is interested to see the arc of how AI has reshaped higher education since late 2022.

Events are organized on two sides: opportunities and frameworks (left) and concerns and disruptions (right). Filter by category using the buttons at the top.

Click any event to expand. Last updated July 24, 2026.

View full screen →

aiX | TZStats aiX Lab | Tian Zheng, Columbia Stats

aiX Weekly — AI in Higher Education (July 29th, 2026)

TL;DR

Table of Contents

This Week at a Glance

Research Highlights

What’s in the News

Most Discussed

Try This Week

The Delayed, Unaided Check (20–30 minutes)

aiX Weekly — AI in Higher Education (July 22nd, 2026)

TL;DR

Table of Contents

This Week at a Glance

Research Highlights

Institutional Movements

What’s in the News

Most Discussed

Interesting Ideas & Repos

What Changed

aiX Weekly — AI in Higher Education (July 15th, 2026)

TL;DR

Table of Contents

This Week at a Glance

Research Highlights

1. Microsoft 2026 AI in Education Report: Universal Adoption, Minimal Training

2. Digital Education Council LATAM Survey: 30,000 Responses Across 29 Institutions

3. HEPI Policy Note: What UK University AI Policies Actually Do

Institutional Movements

1. Lumina-Gallup: 47% of Students Have Considered Switching Majors Due to AI

2. University of Surrey: AI Embedded in Every Degree from September 2026

3. FutureEd: 71 AI Education Bills Across 27 States

What’s in the News

1. Google Unveils Connected AI Tools for Classrooms at ISTE 2026

2. Microsoft Releases Third Annual AI in Education Report at ISTE

3. Times Higher Education: University AI Policies “Promise Support but Deliver Surveillance”

4. AI Campus Index: First National Ranking of Universities on AI Readiness

Most Discussed

1. GPQA Diamond Saturation: AI Models Now 30 Points Above Human PhD Experts

2. Anthropic’s 2026 Agentic Coding Trends Report

3. Brown Professor Suspects Majority of Class Used AI to Cheat on Take-Home Midterm

4. Early-Career Employment Decline in AI-Exposed Occupations

Interesting Ideas & Repos

1. DeepTutor: Agent-Native Personalized Tutoring Platform

2. AI Engineering from Scratch: 503 Lessons, 320 Hours

3. LearnHouse: Open-Source Learning Platform with AI Integration

4. Plano: AI-Native Proxy for Cost Control in Agentic Apps

5. Speechmatics Academy: Open-Source Voice AI Examples and Tutorials

Try This Week

AI Policy Audit: Is Your Syllabus Statement Educating or Surveilling?

aiX Weekly — AI in Higher Education (July 8th, 2026)

TL;DR

Table of Contents

This Week at a Glance

Research Highlights

1. Faster Completion, Less Learning: AI Reduces Study Time and Knowledge Retention

2. AI Grade Inflation Documented Across 500,000 Grades

3. Chinese Study of 26,811 Students: AI Cuts Homework Time, Tanks Exam Performance

Institutional Movements

1. SUNY Adopts Systemwide AI Policy Across All 64 Campuses

2. Entry-Level Job Market Shifts and Higher Education

3. Student AI Bill of Rights Unveiled

What’s in the News

1. Senate Holds First Hearing on AI in K-12 Classrooms

2. NPR/Ipsos Poll: Teachers Say AI’s Impact Will Eclipse the Internet

3. NBC News: Students Use AI “Humanizer” Tools to Beat Detection

4. The Economist: AI Models’ Values Don’t Reflect Most of the World’s

Most Discussed

1. Reddit Study: 270K Posts Map the Evolution of AI Education Discourse

2. August AI Scores 100% on USMLE

3. Students’ Self-Awareness About AI Dependency

Interesting Ideas & Repos

1. FabData-LLM: RAG System for Educational Document Collections

2. Socratic AI Design as Pedagogical Template

Try This Week

aiX Weekly — AI in Higher Education (July 1st, 2026)

TL;DR — Questions This Issue Helps You Think About

Table of Contents

This Week at a Glance

Research Highlights