AI in Higher Education Newsletter

May 22, 2026 · Vol. 18

A weekly brief for the Management Department, Isenberg School of Management, UMass Amherst. By Matthew D. Langenkamp / 雷邁德, prepared in collaboration with Thea 🪻✨.

← Back to archive · ← Back to Dictionary

Executive Summary

Three developments are worth colleagues’ attention this week.

First, the AI verification gap now has numbers. A new Cornell / Berkeley / Science study of more than 95,000 undergraduates at 20 U.S. public research universities found that about one-third of students regularly use generative AI for assignments, and 9% reported using it to cheat. The exact cheating figure should be treated carefully — self-reported misconduct is almost certainly an imperfect measure — but the direction is clear. AI use is now widespread enough that assessment design can no longer assume that polished student artifacts reliably demonstrate independent competence.

Second, the public question about higher education is becoming larger than cheating. Jay Caspian Kang’s recent New Yorker series asks whether AI, demographic pressure, and anti-institutional politics could make college feel obsolete for some families. That is not a trade-press problem. It is a mainstream legitimacy problem. When a general reader encounters higher education through questions about cost, credential value, job-market disruption, and AI, business schools need a better answer than “we are monitoring the situation.”

Third, students increasingly understand AI as career preparation, not just academic-risk territory. CNBC, citing Handshake’s 2026 graduate research, reports that 58% of surveyed college seniors say they need a better understanding of AI to succeed. At the same time, many students say AI has not been meaningfully integrated into their programs. The gap between student expectation and curriculum reality is now part of the employability conversation. For colleagues interested in hearing student perceptions directly, Anthropic also sponsored a roundtable discussion with students in a YouTube format, which may be a useful companion source for this theme.¹

For Isenberg, the practical takeaway is this: AI policy should be connected to evidence of learning and evidence of employability. The question is no longer only “Did the student use AI?” It is: Can the student explain the problem, disclose the tool use, validate the output, defend the judgment, and show competence that survives outside the artifact?

Overview

This week’s lead is the empirical confirmation of a problem we discussed last week. Vol. 17 argued, using AACSB’s recent writing, that generative AI has created a verification gap for business schools. Polished outputs — presentations, memos, portfolios, cover letters, market scans — no longer prove as much as they once did. This week, a new Cornell / Berkeley / Science study gives that concern a much stronger evidence base. More than 95,000 undergraduates across 20 U.S. public research universities were surveyed. The study found widespread generative-AI use, meaningful misuse, and disparities in access and usage. (Cornell Chronicle, May 21, 2026; Berkeley News, May 21, 2026; Science, May 21, 2026)

The second story is the mainstreaming of the legitimacy question. Jay Caspian Kang’s New Yorker newsletter, “Will College Soon Be Obsolete?”, points readers to his multi-part series on whether AI, the enrollment cliff, and anti-establishment politics could change the college decision for today’s children. The importance is not that Kang has discovered a new policy detail. It is that the question has moved into ordinary educated-public discourse: What is college for if AI changes the relationship among learning, credentialing, and work? (The New Yorker, May 21, 2026)

The third story is student preparedness. CNBC’s May 18 piece on AI and the workforce reports that 58% of surveyed college seniors believe they need stronger AI understanding to succeed, drawing on Handshake’s 2026 graduate report. A related April CNBC story reported that only 28% of rising graduates said their schools had meaningfully integrated AI into their programs. That is a gap students will increasingly notice.

AACSB and SUNY remain important, but they should be read this week as background and interpretation rather than repeated headline news. AACSB gives business schools the vocabulary: employability, verification, disclosure, judgment, oral defense, evidence. SUNY gives public systems one governance model. The new material this week is that the evidence base and public legitimacy stakes have sharpened.

1. The Verification Gap Now Has Numbers

Cornell reported this week on a new study, published in Science, analyzing survey responses from more than 95,000 students at 20 U.S. public research universities. The headline finding is straightforward: generative AI is now broadly embedded in undergraduate work. Cornell’s summary says about one-third of students regularly used GenAI tools for assignments, and 9% reported using AI to cheat. The article quotes the researchers’ conclusion directly: “Assessment reform is necessary and urgent.” (Cornell Chronicle, May 21, 2026; Science, May 21, 2026)

Berkeley’s summary of the same study adds useful nuance. About two-thirds of respondents had used GenAI at least sometimes, and almost 40% used it monthly or more. The more frequently students used AI, the more likely they were to report misusing it. At the same time, Berkeley’s write-up cautions against treating prohibition as a complete answer, because AI proficiency is becoming relevant to employment. In other words, students can misuse AI in ways that damage learning, but lack of AI fluency may also damage career preparation. (Berkeley News, May 21, 2026)

A small caution is warranted. The 9% cheating figure is self-reported. Some students who cheated will not say so; some students may define “cheating” differently depending on course rules; and some faculty rules are still unclear. The point is not that 9% is a perfect number. The point is that misuse is measurable at scale, while regular AI use is far larger than the misconduct category. The assessment problem cannot be reduced to catching bad actors.

For business schools, the implication is direct. Traditional artifacts still matter, but they no longer carry the same evidentiary weight. A polished recommendation, slide deck, strategic memo, résumé bullet, market analysis, or written reflection may show something. It no longer shows enough by itself. We need assessment designs that make the student’s reasoning visible.

That means asking for evidence such as:

The practical question is not “How do we ban the tool?” The better question is: What evidence would convince us that the student still owns the judgment?

2. The Public Legitimacy Question Has Moved Mainstream

Jay Caspian Kang’s New Yorker newsletter, “Will College Soon Be Obsolete?”, is worth including not because it is a technical higher-education policy report, but because it captures the new public mood. Kang describes graduation season in Berkeley, then points readers to his recent series on whether today’s children will attend college at all — or whether AI, anti-establishment politics, and demographic shifts will push many families toward other routes. (The New Yorker, May 21, 2026)

That framing matters. The AI-in-higher-education conversation is no longer mainly an internal faculty conversation about cheating and detection. It is increasingly a public conversation about value:

Business schools sit directly in the middle of that question. We do not offer education only as intellectual formation, though that matters. We also offer a professionally oriented credential. Students and families expect the degree to help them enter the labor market. If AI changes early-career work, then our curriculum has to answer that change in visible ways.

The useful move is not panic. College is not about to disappear. But the burden of explanation is rising. A business school should be able to say, in plain language, what students practice here that AI cannot simply replace. Good answers might include problem framing, evidence evaluation, team coordination, client judgment, ethical responsibility, oral defense, implementation under constraints, and the ability to supervise AI rather than merely consume it.

That is also why the Cornell/Berkeley study matters. The public legitimacy question and the assessment-validity question are now linked. If universities cannot show that student work still represents student competence, the external value of the credential weakens.

3. Students Are Asking for Career Preparation, Not Just Permission Rules

CNBC reported this week that, as companies integrate AI into work, new graduates across industries will need experience using AI tools. Citing Handshake’s 2026 graduate research, CNBC reports that 58% of surveyed college seniors say they need a better understanding of AI to succeed. A related CNBC report in April said that only 28% of rising graduates believed their school had meaningfully integrated AI into their programs. (CNBC, May 18, 2026; CNBC, Apr. 29, 2026)

That mismatch should get our attention. Students are not experiencing AI only as a temptation to cheat. They are also experiencing it as a workplace expectation. Some are anxious that their majors may become less valuable. Some are experimenting with tools on their own. Some are probably receiving inconsistent messages from different courses: use it here, never use it there, disclose it in one class, pretend it does not exist in another.

For a business school, the student question is practical: Will this degree help me become employable in an AI-mediated workplace?

That does not mean every course should become an AI-tools course. It means courses should be clearer about the relationship between AI and the learning objective. Sometimes unaided work is necessary because students need to build foundational skill. Sometimes AI-assisted work is appropriate because the professional world now works that way. Sometimes students should use AI and then critique it, because judgment is the point of the exercise.

A useful course-level vocabulary might look like this:

  1. No-AI work — because the assignment is designed to build unaided fluency.
  2. AI-assisted work — because the tool can support drafting, exploration, or analysis, but the student must disclose use and own the final judgment.
  3. AI-required work — because the learning objective is to practice supervising, validating, or improving AI output.

The key is not uniformity for its own sake. The key is clarity. Students should not have to infer the AI policy from faculty mood.

4. The Experience Gap: When First Jobs No Longer Teach First Skills

A related Fortune piece argues that AI is beginning to erode the traditional first rung of career learning. If AI automates some of the routine tasks that once defined entry-level work, students may lose part of the pathway through which young professionals learned judgment by doing. The author’s proposed response is that higher education should embed more real-world experience into the curriculum before students graduate: simulations, projects, applied work, and other forms of practice. (Fortune, May 15, 2026)

This argument is especially relevant for business education. Much of early professional development has historically happened through supervised low-stakes work: assembling data, drafting first passes, preparing slides, doing customer or market research, writing summaries, building simple models, sitting in meetings, revising after feedback. If AI absorbs some of that work, then the curriculum has to provide more of the developmental ladder.

That makes applied, evidence-rich assignments more important, not less:

The risk is that students graduate with more polished outputs and less practiced judgment. The opportunity is to make judgment itself the object of assessment.

5. AI-First Degree Models and the Competitive Edge of Traditional Programs

The Chronicle of Higher Education also reported this month on a new “AI First” college model aiming to offer cheaper, employer-friendly degrees, with Sal Khan involved and major technology firms such as Microsoft and Google having input into what is taught. Details are still emerging, and the item should be treated as a watchlist signal rather than a settled institutional model. (Chronicle of Higher Education, May 2026)

Still, the signal is worth noting. Traditional universities are no longer competing only with peer institutions. They are also competing with alternative credential structures that promise lower cost, tighter employer alignment, and faster curriculum adaptation. Some of those models will overpromise. Some will fail. But they will still shape student and family expectations.

For Isenberg, the response is not to imitate every AI-first model. The response is to be explicit about what a serious business-school education provides that a thin credential cannot: structured progression, faculty judgment, peer learning, accountable teamwork, ethical framing, analytical depth, professional standards, and repeated practice under constraint.

The traditional degree is strongest when it can show its work. That phrase applies to institutions as much as students.

6. Governance: AACSB and SUNY Are Background Now, Not the Headline

Last week’s issue gave substantial attention to AACSB and SUNY, so they should sit lower this week. But they still provide the interpretive frame.

AACSB’s April piece, “Bridging the AI Employability Gap,” argued that AI has weakened traditional employability signals and that business schools need stronger evidence of how students frame problems, manage tradeoffs, validate AI outputs, and defend outcomes. AACSB’s May piece, “AI Integration, Not Prohibition,” reinforces the same direction: disclosure, explanation, human judgment, and structured integration rather than blanket bans. (AACSB Insights, Apr. 22 and May 2026)

SUNY’s systemwide AI policy remains the clearest public-system governance example. It matters because it shows AI moving from individual experimentation toward systemwide roles, responsibilities, privacy concerns, and learning expectations across 64 campuses. But this week, it should be treated as background context rather than repeated news. (Inside Higher Ed, May 4, 2026)

The newer governance item is broader state attention. FutureEd’s 2026 tracker of state AI-in-education bills shows that AI policy is moving beyond campus committees into legislatures. Much of that activity is K–12, but public higher education will not be insulated from the same pressures: privacy, procurement, academic integrity, literacy, bias, and accountability. (FutureEd, May 2026)

The department-level implication is modest but important. Before large policy systems settle, departments can still do useful work:

7. Practical Implications for Isenberg

The through-line this week is that AI has made the evidence problem more visible. Students are using the tools. Employers increasingly expect AI fluency. Public writers are questioning the value of college. Researchers now have large-scale evidence of both use and misuse. Accreditors are asking business schools to think in terms of capability and employability.

That leaves us with a practical agenda.

First, assignment design should separate artifact quality from demonstrated competence. A good memo is not enough. A good memo plus process evidence is stronger.

Second, AI-use disclosure should become normal professional documentation, not a confession. Students should be able to say what they used, why they used it, what they checked, and where they exercised judgment.

Third, oral defense and in-class explanation should return as serious assessment tools. If a student cannot explain the recommendation, the spreadsheet, the market claim, or the strategic logic, the artifact should not receive full evidentiary credit.

Fourth, we should preserve unaided practice where it matters. Integration is not the same as surrender. Some skills still need to be built without assistance before students can responsibly supervise a tool.

Finally, we should describe our value proposition in terms of judgment. AI can produce language. It can produce plausible analysis. It can accelerate research. But it does not replace accountable professional judgment under constraint. That is where business education should plant its flag.

Closing Thought

Last week’s question was: What evidence would show that an Isenberg student can use AI well without outsourcing judgment and understanding to it?

This week’s answer is sharper: we need assessments that make judgment visible. Not because AI is a passing disruption, and not because every student is cheating, but because the old evidence system was built for a world in which producing the artifact was much closer to proving the competence.

That world has changed.

Questions or topics for next week? Reply to mlangenkamp@umass.edu. Prepared by Thea 🪻✨

Source Notes

¹ Anthropic-sponsored student roundtable, YouTube: https://youtu.be/N5yJJA0NCU0?si=7gzowF9n_fAxylNb

Return to Dictionary All Entries (A–Z) For Students Other Writing