AI & Higher Education Global Brief: Evidence Before Expansion | October 11, 2026

The University of Maryland built its own AI tutor, tested it in a randomized trial before rolling it out, and found that students offered access earned lower grades. Then it paused. That decision frames this week. Universities keep expanding AI tutoring, integrity tools, research access, and partnerships, while the evidence that these investments improve learning lags behind, and institutions are still working out how to protect student data and keep human judgment in consequential decisions. The question running through every story from October 4 to 11: are institutions evaluating the educational consequences of AI as carefully as they evaluate its capabilities?

“We don’t have clear findings from the study.”
Jennifer King Rice, Senior Vice President and Provost, University of Maryland (Palmer, 2026)

More British universities walk away from Turnitin

The Details

Newcastle University will stop using Turnitin after the 2026-27 academic year, citing proposed changes to the company’s end-user license agreement and the potential use of staff and student data for AI purposes. The University of Westminster will retire the service when its contract ends in July 2027, calling the plagiarism checker largely ineffective against generative AI misuse and its AI detectors “probabilistic rather than conclusive.” The University of Southampton announced earlier this year that it will end its contract. Aberdeen is reviewing its use, and Cambridge says it will not accept license changes that affect data privacy or intellectual property. Commentators also point to cost; Southampton reported paying £87,621 for one year. Turnitin says it does not train generative AI on student work, and it has postponed the proposed license changes to September 2027 while it works with a university representation group (Luckhurst, 2026).

Why it Matters

Similarity checking was built to catch copied text. Generative AI produces new text, so the core problem has changed. The question is not which detector to buy next. Institutions need assessments that generate evidence of student understanding, fair procedures for investigating suspected misconduct, and clear protection of student work. Every integrity contract up for renewal deserves a review of what the tool can reliably establish, what it costs, and what it does with student data.

Grant proposals screened out by AI will get a human review after all

The Details

CRANE, a UK cybersecurity research network hosted at the University of Oxford and funded by the Engineering and Physical Sciences Research Council, part of UK Research and Innovation, used AI to triage 179 proposals in its phase one pilot. Almost half were eliminated without human review. After criticism from researchers, CRANE apologized this week for not being transparent about its AI triage. It will re-review every application the AI screened out, reconvene its funding panel, and suspend offers already issued (Jones, 2026a).

Why it Matters

Disclosure alone would not have fixed this. When AI screens out applicants before any person looks at them, the process has no point at which a reviewer can catch an error. Universities and funders need to decide, in advance, where AI may assist reviewers and where decisions about people’s opportunities require direct human evaluation.

Policy & Governance
  • Data governance gaps are slowing AI, EDUCAUSE session finds

    At the EDUCAUSE Annual Conference in Denver, four university data officers shared a year of survey work and discussion from the EDUCAUSE data governance community group. Of 65 respondents, 27 said data responsibilities at their institutions were unclear, and 18 said no one was in charge of monitoring data quality. The presenters stressed that terms such as “student” or “enrollment” mean different things to institutional research, finance, and the registrar. At Rowan University, a chatbot asked to “show me enrollment” counted every academic year and reported 600,000 students (Sourwine, 2026a). An AI system can return data that is technically correct and still give the wrong answer. Before AI supports advising, enrollment, finance, or planning decisions, institutions need shared definitions, named data owners, access rules, and a process for correcting errors.

  • Copenhagen puts an anthropologist in charge of AI strategy

    Morten Axel Pedersen became the University of Copenhagen’s first deputy prorector for AI in September. The university has invested DKK 110 million (about £13 million) in an AI office, a cross-university task force, AI labs, a university generative AI assistant, and an “AI traffic light” showing which data can be used with which systems. Over the next year, Pedersen plans to review every course to decide whether AI should be included, excluded, or required, with each discipline setting its own limits. He argues that leaders must observe how people actually use AI, because much of that use currently “takes place in the shadows” (Asaf, 2026). AI strategy is as much about behavior, disciplinary norms, and organizational culture as it is about technology.

Programs, Research & Infrastructure
  • EDUCAUSE researcher calls for baseline AI skills in every discipline

    Nicole Muscanell, senior researcher at EDUCAUSE, pointed to the 2026 Students and Technology Report, in which 53% of students said their courses had restrictive AI policies, even as employers signal growing demand for AI skills. She recommends base-level competence suited to each field: using AI-enabled tools, navigating their interfaces, and evaluating outputs. Most students do not need to understand the science behind large language models. She also notes that many instructors have not had the time to integrate AI thoughtfully, given staffing shortages and heavier workloads (Sourwine, 2026b). Programs should define the AI capabilities their graduates actually need rather than assuming every student needs advanced technical training.

  • Colleges expand AI security education

    Carnegie Mellon University, the University of West Florida, Georgia Tech, Ferris State University, and others are building or expanding AI security courses, certificates, and degrees. Carnegie Mellon’s AI security master’s, launched in 2022, now draws almost as many applications as its information security master’s, launched in 2003. A Pentagon-funded consortium of roughly three dozen universities led by the University of West Florida is developing new AI security career pathways (Geller, 2026). As agentic systems take on more tasks, securing them becomes a workforce need in its own right, and technical programs should treat AI security and governance as core content.

  • AI-supported case learning still depends on instructional design

    A George Mason University scoping review led by John Cantiello, with Renee Geschke, examined 83 articles published from 2017 through 2025 across health administration, nursing, public health, medicine, and business. Case-based learning was linked to gains in critical thinking, decision-making, teamwork, and engagement, but the authors stress that “technology alone is not enough.” Effective cases need clear goals, structure, timely feedback, and focused discussion. Generative AI can omit information or give inaccurate answers, and students may lean on it too heavily (Pearce, 2026). Institutions adopting AI-supported simulations need instructors prepared to design them and to judge whether students are building the intended capabilities.

Other
  • Publishers open more than 20 million papers to AI agents

    Six publishers, Sage, BMJ Group, IOP Publishing, IGI Global Scientific Publishing, RCNi, and Wanfang Data, have agreed to provide AI agents with pay-per-use access to more than 20 million research papers through Redpine, a Swedish start-up founded in 2024. Each publisher sets what an agent can use and the price per transaction, and Redpine says authors will be paid. The company argues that retrieving and citing primary literature at the moment of answering reduces errors, a claim not yet independently tested (Jones, 2026b). For research universities and libraries, AI agents may increasingly sit between researchers and the literature they once retrieved and evaluated themselves, raising questions about licensing, access, and verification.

Do It Now Checklist

This Week

On Your Radar

Betting On: Evidence-Based Implementation

This week’s stories share one demand: show that AI improves the work it is meant to support. Maryland tested its tutor and paused when the results raised questions. British universities are asking what their integrity tools can still establish. A research network reversed course after AI screened out applicants with no human in the loop. Readiness cannot be measured by technology access, adoption rates, or vendor commitments alone. Responsible implementation requires clear objectives, human oversight, faculty preparation, and credible measures of educational value.

With Inspiration Moments, we share motivational nuggets to empower you to make meaningful choices for a more fulfilling future. Remember: “Life happens for you, not to you, to live your purpose.”

Respectfully,
Dr. Lynn Austin

Leave a Reply

This site uses Akismet to reduce spam. Learn how your comment data is processed.

Back to top