University Leadership Brief

Executive Summary

Leadership Briefing: The Vendor Deal Is the Policy

While your cabinet deliberates an AI strategy, peer institutions are signing the precedent for you. Cal State’s system-wide OpenAI deployment is now being publicly refused by some of its own students and faculty Cal State struck a deal with OpenAI. Some students and …. ASU faculty are pushing back on an institutional AI Course Builder rolled out without their input Faculty Concerned About ASU’s New AI Course Builder. Surrey has committed to embedding AI in every degree from September 2026 Surrey embeds AI in every degree from 2026. None of these decisions has settled — all of them are now templates your board will be asked to copy.

The strategic challenge. The decision in front of you is not whether to adopt AI. It is whether to let an enterprise license function as your AI policy. When OpenAI’s Edu terms, a vendor’s course-builder defaults, or a system-wide MOU set the operational rules — what counts as legitimate use, what data flows where, what faculty can refuse — shared governance has been outsourced to a procurement document. The South African government has already demonstrated the downstream cost of skipping the verification layer: its national AI policy cited fabricated, AI-generated research South Africa’s AI policy cited fake research, created by AI. The Adelphi suit shows the parallel risk on the enforcement side: institutions accusing students of AI use without defensible process now face litigation An Adelphi University student was accused of using AI to … - Newsday.

What this briefing provides. Drawn from this week’s 6,252 sources: the policy framework options peer institutions are actually testing, the documented failure patterns — vendor capture, due-process exposure, fabricated evidence in policy itself — to design against, and the governance and resource implications your provost, general counsel, and faculty senate need before the next contract cycle closes the question for you.

Critical Tension

The Strategic Dilemma

The institutional question this week is no longer whether to deploy AI but on whose terms. Cal State has signed a system-wide deal with OpenAI that students and faculty are now publicly refusing to use Cal State struck a deal with OpenAI. Some students and …. Arizona State’s new AI Course Builder has triggered formal faculty objections about who actually authors a course Faculty Concerned About ASU’s New AI Course Builder. Surrey has gone further still, announcing AI will be embedded in every degree from September 2026 Surrey embeds AI in every degree from 2026. The strategic tension is between scaling vendor-mediated AI access fast enough to keep institutional relevance versus preserving the governance, pedagogical authority, and legitimacy that make a credential worth issuing. Both halves are existential. Sitting still loses recruitment and operating-cost arguments; moving fast surrenders shared governance to a procurement contract.

This is not solvable by more data — it is a values allocation problem dressed as an IT decision. The peer-reviewed work calling AI a “policy response to higher education in crisis” is explicit that retention dashboards and course-builder tools are being adopted because enrollment math is bad, not because pedagogical evidence is in Risk, Retention, and the Algorithmic Institution: Artificial Intelligence as a Policy Response to Higher Education in Crisis. Difficulty here is hard: the evidence base lags the contracts by years, and the contracts are being signed now.

Why Peer Institutions Aren’t Helping

The sector is not converging — it is splintering, and the splinters carry hidden liabilities. South Africa’s national AI policy was found to cite fake research generated by AI itself, a governance failure at the document level South Africa’s AI policy cited fake research, created by AI. Adelphi University is now in active litigation after accusing a student of AI use on contested detector evidence Adelphi University accused a student of using AI to plagiarize. He …, part of a documented wave of detection-based lawsuits with poor institutional outcomes AI Detection Lawsuits: Every Student Case, Outcome, and What the Data …. Students at Staffordshire have already organized against courses they perceived as AI-taught 'We could have asked ChatGPT': students fight back over course taught by AI.

The pattern is consistent: institutions that adopted detection-first policies are losing in court; institutions that adopted vendor-embedded curricula are losing the consent of the governed; institutions that wrote ethics-forward policies without source verification are publishing fabricated citations. Copying any of these postures imports the failure mode with it. There is no safe peer to mimic — there are only different lawsuits.

What Complicates Navigation

The voices most affected by these decisions are the least represented in the discourse shaping them. Across the 2,287 higher-education articles surfaced this week (out of 6,252 total), student perspectives appear in only 3.76% of coverage; parents register at 0.29%, organized critics at 0.29%, and — tellingly — vendors at 0.29%, because vendors do not need to argue in public when they are already inside the procurement process. Leadership is making system-wide decisions inside a discourse where the people who will be graded by the system, pay for it, or audit it are functionally absent.

The dominant framing — AI as a productivity tool embedded in the institution’s existing workflows — performs specific erasures. It hides that ChatGPT Edu is a product line with a roadmap OpenAI controls ChatGPT Edu at OpenAI - OpenAI Help Center, that “designing smarter classes” reframes course design as prompt engineering #AnteaterIntelligence: Designing Smarter Classes with ZotGPT, and that the labor-market argument for embedding AI everywhere — “students need exposure” — coexists with Yale CELI evidence that the entry-level rungs those students would climb are being cut AI won’t kill your job — it will kill the path to your first one. MIT Sloan’s work on generative AI “persuasion bombing” should sit on every cabinet’s desk before the next vendor demo: these systems are documented to shift user judgment in measurable ways How generative AI 'persuasion bombs' users, which means the procurement conversation itself is being conducted with a counterparty optimized to win it.

The governance question is not what AI policy to adopt. It is which constituencies have standing in the room when the contract is signed.

Actionable Recommendations

Leadership Briefing: Five Moves to Make Before the Next Procurement Cycle

Week: . Total sources surveyed: 6252.

The decisions in front of cabinets and provosts this spring are not, despite the marketing, about “AI strategy.” They are about procurement, shared governance, assessment policy, and legal exposure — four areas where institutional reflexes are pulling administrations toward fast, visible commitments that the evidence already shows are failing. What follows is five moves, each built around a specific failure that other institutions have already run.

1. Stop signing campus-wide enterprise deals before faculty senate has seen the contract

The common move — a presidential announcement that the university has “partnered” with OpenAI or another frontier-model vendor for a campus license — is failing in a specific way. At Cal State, the system signed a $16.9M ChatGPT Edu deal and is now contending with students and faculty who refuse to use the tool, citing labor displacement, environmental cost, and the lack of academic input into the contract Cal State struck a deal with OpenAI. Some students and …. At Arizona State, faculty are publicly resisting an AI course-builder rolled out without curricular review Faculty Concerned About ASU’s New AI Course Builder. The hidden complexity: vendor terms (ChatGPT Edu at OpenAI - OpenAI Help Center) define data flows, audit rights, and model-update cadence — all of which structure pedagogical decisions for years afterward.

Recommended alternative: route any system-level model contract through faculty senate and a student-government review before signature, with a published memo on data, IP, and termination rights.

Implementation framework: - Phase 1 (Month 1–2): inventory all existing AI vendor contracts; redline against five disclosure standards (training-data use, retention, audit, exit, price-lock). - Phase 2 (Month 3–4): convene a joint senate–IT–general-counsel committee to publish a procurement standard. Require any deal above a threshold (e.g., $500K or campus-wide deployment) to go through it. - Phase 3 (semester end): publish all signed AI vendor terms internally; report annually.

Required resources: 0.25 FTE counsel, 0.1 FTE faculty governance liaison, no new software. Success metrics: zero contracts above threshold signed without senate review; published vendor register. Risk mitigation: watch for “pilot” framing used to bypass governance — pilots that touch grading or admissions are not pilots.

2. Replace AI-detection enforcement with assessment redesign

The reflex on academic integrity has been to license a detector and discipline accordingly. That posture is now generating active legal exposure. Adelphi University is being sued by a student who says he was wrongly accused on the basis of detector output An Adelphi University student was accused of using AI to … - Newsday, Adelphi University accused a student of using AI to plagiarize. He …; and the broader litigation pattern is now traceable AI Detection Lawsuits: Every Student Case, Outcome, and What the Data …. The detection-arms-race itself is documented: students are now running their work through “humanizers” specifically to defeat detectors To avoid accusations of AI cheating, college students turn to AI - NBC News. The methodological case against detection-based adjudication is also clear in the literature Contra generative AI detection in higher education assessments.

Recommended alternative: shift the locus of integrity from detection to assessment design — oral defenses, in-class drafting, process portfolios, supervised revisions. The literature on authentic assessment is now substantial enough to anchor policy Beyond Detection: Redesigning Authentic Assessment in an AI … - MDPI, Reimagining Writing Assessment for the AI Era: A Systematic Review on Balancing AI Support and Authentic Skill Growth, PDF Authentic Assessment in the Age of AI - marcbowles.com.

Implementation framework: - Phase 1 (Month 1–2): freeze new detector licenses; draft an academic-integrity policy that bars detector output as sole evidence. - Phase 2 (Month 3–4): fund a teaching-and-learning center cohort (15–25 faculty) on authentic assessment redesign, with course-release stipends. - Phase 3 (next assessment cycle): require redesigned assessment plans for high-enrollment writing-intensive courses; track Title IX–adjacent due-process complaints.

Required resources: ~$150K stipend pool; 1 FTE instructional designer; legal review of integrity policy. Success metrics: integrity cases adjudicated on process evidence (drafts, conferences) rather than detector scores; reduction in formal grievances. Risk mitigation: detector output appearing as “supporting” evidence is still detector output — bar it explicitly.

3. Treat AI-generated content in your own institutional documents as a compliance risk

Leadership often assumes the AI risk is downstream — students cheating, faculty over-relying. The South African case is a warning shot: a national AI policy was found to cite fabricated AI-generated references South Africa’s AI policy cited fake research, created by AI. Universities producing strategic plans, accreditation self-studies, and Title IX investigations on accelerated timelines are running the same risk.

Recommended alternative: an institutional verification protocol for any AI-assisted document leaving the institution.

Implementation framework: - Phase 1 (Month 1–2): require disclosure of AI use in any board-, accreditor-, or regulator-facing document. - Phase 2 (Month 3–4): designate citation-verification responsibility (typically the office of institutional research) for AI-assisted documents. - Phase 3 (semester end): audit a sample of AI-touched institutional documents for hallucinated citations; publish findings internally.

Required resources: 0.25 FTE in IR; existing reference-checking workflows. Success metrics: zero fabricated citations in regulator-facing documents; disclosure rate. Risk mitigation: an accreditation self-study with hallucinated sources is an accreditation crisis. Treat it as one before it happens.

4. Build the curricular response to entry-level labor collapse — not the press-release response

The labor signal this quarter is sharp: the Yale CELI analysis argues that AI is not eliminating jobs broadly but is hollowing out the entry-level rung that graduates step onto AI won’t kill your job — it will kill the path to your first one. This is a placement and tuition-revenue problem before it is a curricular one. The standard reflex — announcing that “every student will graduate AI-literate,” as Surrey has done Surrey embeds AI in every degree from 2026 — is the easy half. The harder half is that the skills demanded are shifting from production to judgment L’IA sait tout produire… mais pas encore juger, and that algorithmic hiring tools are themselves discriminatory in documented ways Utiliser l’IA pour recruter ? Attention aux risques de ….

Recommended alternative: a vertically integrated curricular redesign that pairs AI fluency with judgment-heavy capstone work — the After shock model of cross-disciplinary, project-based integration is the relevant precedent.

Implementation framework: - Phase 1 (Month 1–2): convene deans plus career services to map degree-by-degree exposure of the entry-level pipeline. - Phase 2 (Month 3–4): fund 5–10 vertically integrated capstone redesigns; build employer-side data collection on graduate hiring. - Phase 3 (one-year cycle): publish placement data disaggregated by AI-exposure of target roles.

Required resources: $500K–$1M curriculum-redesign pool; existing career-services infrastructure. Success metrics: placement rates in AI-exposed fields; employer satisfaction with graduate judgment, not just tool fluency. Risk mitigation: “AI literacy” as a one-credit gen-ed module is the failed obvious approach — it neither moves placement nor satisfies employers.

5. Govern AI-in-grading and AI-in-retention separately from AI-in-teaching

Faculty using ChatGPT to plan a class #AnteaterIntelligence: Designing Smarter Classes with ZotGPT is a different governance question from an institution deploying AI to grade student work Is It Ethical to Use AI to Grade? - Education Week, and both are different from algorithmic retention systems that route students based on risk scores Risk, Retention, and the Algorithmic Institution: Artificial Intelligence as a Policy Response to Higher Education in Crisis. Treating them under one “AI policy” obscures where the consequential decisions sit.

Recommended alternative: a tiered governance framework that distinguishes pedagogical use (faculty discretion + disclosure), evaluative use (requires due-process review), and predictive/managerial use (requires senate, IRB-equivalent, and student-representative approval). The general scaffolding is consistent with emerging governance literature AI Leadership in Education: A Governance Framework to Scale Safely.

Implementation framework: - Phase 1 (Month 1–2): inventory all current AI uses by tier. - Phase 2 (Month 3–4): publish tier definitions and approval pathways; require explicit recategorization when a tool’s use changes tier. - Phase 3 (annual): public-facing AI use register, by tier.

Required resources: 0.5 FTE policy lead; existing IRB and senate structures. Success metrics: every consequential AI use traceable to an approval; no tier-2 or tier-3 use deployed without it. Risk mitigation: vendors will frame retention or grading tools as “decision support” to dodge the higher tier. The tier is determined by where the decision actually lands, not by how the vendor describes it.

These five moves share a posture: the consequential AI decisions on campus are not technical, they are governance decisions made under procurement and timeline pressure. The institutions that will look competent in three years are the ones whose contracts, integrity policies, and curricular responses survive contact with their own faculty, students, and accreditors.

Supporting Evidence

The Evidence Base: What Leadership Can and Cannot Conclude

Evidence Landscape

This week’s corpus comprises 6,252 articles, of which 2,287 sit in the higher-education category. The evidentiary quality is uneven in a way leadership should name explicitly: vendor-adjacent material (OpenAI’s own product documentation for ChatGPT Edu at OpenAI, institutional press releases like Surrey embeds AI in every degree from 2026) sits alongside peer-reviewed work on assessment redesign (Reimagining Writing Assessment for the AI Era, Beyond Detection: Redesigning Authentic Assessment in an AI-Mediated World) and adversarial reporting (Cal State struck a deal with OpenAI. Some students and faculty refuse to use it, Faculty Concerned About ASU’s New AI Course Builder).

What this evidence can tell you: it documents adoption patterns, governance failures, and the contour of faculty resistance. What it cannot tell you: longitudinal learning outcomes, the actual cost-benefit of system-wide vendor licenses after the introductory pricing window closes, or what happens to the entry-level labor market your graduates depend on (AI won’t kill your job — it will kill the path to your first one).

Stakeholder Perspective Gaps

The corpus this week surfaces no quantified missing-perspectives data, but the qualitative absence is loud: students who refused the Cal State OpenAI deployment are quoted in trade press, not consulted in the procurement decision itself (Cal State struck a deal with OpenAI). Faculty at ASU learned about the AI course builder through announcements rather than shared governance (Faculty Concerned About ASU’s New AI Course Builder). When implementation precedes consultation, downstream legitimacy erodes — and the legitimacy deficit shows up later as grievances, AAUP letters, and the kind of student lawsuits already accumulating in AI Detection Lawsuits: Every Student Case, Outcome, and What the Data Shows.

Documented Failure Patterns

Three failure categories are well-attested in this week’s evidence. First, epistemic failures in policy itself: South Africa’s national AI policy cited fabricated, AI-generated references (South Africa’s AI policy cited fake research, created by AI) — a cautionary tale for any institution drafting AI strategy with AI assistance. Second, detection-regime failures: the Adelphi case (An Adelphi University student was accused of using AI, Adelphi University accused a student of using AI to plagiarize) joins a documented pattern of false-positive accusations driving litigation, and the academic literature already argues against this enforcement posture (Contra generative AI detection in higher education assessments). Third, pedagogical-substitution failures: the Staffordshire course “taught in large part by AI” produced student revolt (We could have asked ChatGPT).

The risk-management read: enforcement-first AI strategies generate measurable legal exposure; substitution-first strategies generate measurable enrollment and reputational exposure.

Power and Framing Analysis

The dominant frame in vendor-supplied material is “tool” — AI as neutral instrument the institution chooses to deploy. The frame obscures that procurement decisions at Cal State, ASU, and Surrey hand pedagogical authority to a vendor’s roadmap. When OpenAI ships a model update, your syllabus changes. The MIT Sloan reporting on generative AI ‘persuasion bombs’ and the Conversation France piece on AI as production-without-judgment (L’IA sait tout produire… mais pas encore juger) both point at the same thing from different angles: the “tool” frame hides who is shaping cognition and on whose terms.

Research Gaps Affecting Strategy

Leadership lacks rigorous evidence on four questions you will be asked to decide anyway: (1) what the five-year retention effect of an institution-wide vendor lock-in looks like (Risk, Retention, and the Algorithmic Institution raises the question without resolving it); (2) whether AI-augmented grading preserves assessment validity (Is It Ethical to Use AI to Grade?); (3) what authentic assessment costs in faculty time at scale (Authentic Assessment in the Age of AI); (4) whether “AI literacy” as currently operationalized produces the judgment graduates need or merely fluency with one vendor’s interface.

Secondary Tensions

Beyond the primary cheating-vs-pedagogy tension, three secondary contradictions deserve naming. Equity claims for AI access run against documented bias in deployment (Utiliser l’IA pour recruter ? Attention aux risques de discriminations) and against linguistic exclusion (Parler à l’IA en luxembourgeois, un défi encore loin d’être gagné). Governance frameworks (AI Leadership in Education: A Governance Framework to Scale Safely) presume a stable model landscape that the quarterly release cycle does not provide. And the integrity-vs-innovation framing (The AI Dilemma: When Innovation Outpaces Integrity) is a values question, not an optimization problem — it cannot be resolved by procurement.