Higher Education & ResearchIssue 494 records
Student Success
For 2026-10-22, student-success signals are the gap between AI access and student support, and calls to gate AI uses on evidence. The Harvard Crimson (22 Oct) reports students using FAS-provided Claude, most purposefully where courses embed it. Idaho State (22 Oct) won a $300,000 grant for a locally hosted Canvas writing assistant gated by a benchmark, bias audit and accessibility review. On scrutiny, a National Disability Center report (October 2026) finds over 90% of surveyed students learned AI alone and only 6% of disabled students got instructor training, and APA and Springfield College learning scientists (Inside Higher Ed, 22 Oct) urge small studies testing transfer without AI. Operating takeaway (interpretation): pair access with training and course embedding, and gate student-facing tools on learning-goal checks. Limits: reporting, plans, opinion and an exploratory convenience survey; no controlled study dated 21 to 22 Oct.
- Evidence records
- 4
- Cross-source patterns
- 2
- Evidence classes
- 2 independent reporting1 vendor claim1 academic research
- Outcomes
- 2 emerging1 cautionary1 mixed
- Source freshness
- 3 new this fortnight1 undated
- Research completed
- 2026-10-27
Choose a role to see its takeaway beside every record in the ledger.
Patterns across the evidence
Synthesis · Lighthouse Advisory interpretation
Institutions are widening AI access faster than they are training students to use it, and uptake and equity depend on that gap
The National Disability Center report (October 2026) finds over 90% of surveyed students learned AI on their own and only 6% of disabled students received instructor AI training, while cost and course restrictions each limited over 40% of students. The Harvard Crimson (22 Oct) describes heavier academic use where courses embed Claude in assignments. Interpretation: access alone yields partial and uneven use; course embedding and training appear to be where adoption is decided, though neither source measures learning outcomes.
Operating questionFor each institution-wide AI tool, what share of students has received training, how is active use counted by student group, and which courses embed it in assignments?
Supporting evidenceNational Disability Center for Student SuccessHarvard University Faculty of Arts and Sciences
Gate student-facing AI on learning goals and checks before scaling it
Idaho State (22 Oct) plans to gate its Canvas writing assistant behind an internal benchmark, a bias audit, an accessibility review and semesterly faculty sampling, with referrals to the Writing Center, before expanding beyond a first cohort. APA and Springfield College learning scientists (Inside Higher Ed, 22 Oct) argue each AI use should be judged by whether it strengthens or replaces student thinking, with small iterative studies that test transfer without AI and compare student groups. Interpretation: both point to evaluating specific uses before scale; neither provides outcome data, as ISU's tools are not yet deployed and the essay is opinion.
Operating questionBefore a student-facing AI tool expands, what benchmark, accessibility and bias checks, and AI-free transfer measure must it pass, and who signs off?
Supporting evidenceIdaho State UniversityAmerican Psychological Association; Springfield College (authors)
Evidence ledger
Full record · every source keeps its link and limitations
Idaho State University wins a $300,000 grant to expand its locally hosted BengalBot AI, adding a Canvas writing assistant with Writing Center referrals, a bias audit and accessibility review before wider student use
On 22 October Idaho State University reported a $300,000, two-year Accelerate Idaho AI Innovation Award to expand BengalBot, its locally hosted AI system. The grant funds three efforts: embedding BengalBot in the Applied AI Certificate and new AI sciences bachelor's degree (50 to 75 students a year building and auditing AI tools); a Canvas writing assistant that gives feedback on structure and grammar rather than writing papers, with built-in referrals to the ISU Writing Center; and a Stage Hub to reduce administrative load on student stage managers. Per the grant application, data stays on ISU-controlled servers, the writing assistant gets an accessibility review and bias audit before expanding beyond its first group, an English professor reviews a sample of its feedback each semester, and no tool reaches students without passing an internal benchmark. About half the funding buys a server maintained by College of Technology students; code and curricula will be openly licensed, and a theatre pilot and limited English-class use are expected by fall 2027.
Why it matters, evidence and limitations
- Why it matters
- A concrete pre-launch control set for a student-facing writing assistant (human referral, faculty sampling, bias and accessibility review, benchmark gate) that other public institutions can adapt before scaling AI writing help.
- Evidence and measured results
- Institutional description of grant scope and planned safeguards. No measured outcomes; tools are not yet deployed to students.
- Limitations and uncertainty
- Announcement of funded intent; no deployment, usage or learning outcomes. Safeguards come from a grant application. Single regional public university.
Harvard students use FAS-provided Claude for coursework, embedded course assignments and personal projects after the FAS moved ChatGPT Edu to request-only access
The Harvard Crimson reported on 22 October that Harvard College students are using institutional access to Anthropic's Claude, provided by the Faculty of Arts and Sciences from this summer, for coursework, extracurricular projects and personal tasks. The FAS announced in April that it would add Claude while scaling back ChatGPT Edu to request-only access. Some courses embed Claude in assignments, including a Gen Ed course on NeuroAI and a first-year seminar taught by the FAS senior adviser on generative AI, a former Anthropic employee. One student said there is no limit on credits. Students described building personal apps and research scaffolding, and some said vendor values shaped their preference for Claude.
Why it matters, evidence and limitations
- Why it matters
- Shows how an institution switching its default AI vendor plays out in student practice, and that syllabus-embedded uses drive academic adoption; relevant to U.S. campuses deciding platform portfolios.
- Evidence and measured results
- Student interviews reported by a student newspaper. No measured outcomes or usage data.
- Limitations and uncertainty
- Small, self-selected interview set; one student claim that credits are unlimited is not confirmed by the FAS in the article. No outcome or cost data.
Inside Higher Ed opinion: APA and Springfield College learning scientists argue the science of learning should guide AI use, and call for small iterative studies testing transfer without AI
In a 22 October Inside Higher Ed opinion essay, Beth M. Schwartz (APA senior director of teaching and learning) and Christopher Hakala (Springfield College) accept that evidence on whether generative AI improves student learning remains limited and that some studies suggest unguided use could undermine it. They argue that the established science of learning still gives a framework: learning requires targeted cognitive effort, so the question is whether a specific AI use strengthens or replaces that effort. They propose that faculty start from a learning goal and ask which cognitive work must remain with the student and whether students can demonstrate learning independently. They recommend that institutions support smaller, iterative studies of specific AI uses, comparing student work before and after AI assistance, testing transfer without AI and comparing results across disciplines and student populations.
Why it matters, evidence and limitations
- Why it matters
- Gives student-success leaders an evaluation frame for AI pilots that centres independent transfer and subgroup differences rather than tool adoption, relevant given recent mixed course-level trial results.
- Evidence and measured results
- Expert opinion grounded in prior learning-science literature; no new data.
- Limitations and uncertainty
- Opinion, not research; the authors acknowledge the AI-specific evidence base is limited. Recommendations are not tested here.
National Disability Center for Student Success report: over 90% of students learned AI on their own, 82 to 92% of disabled students said AI met access needs, but only 6% got instructor AI training
The IES-funded National Disability Center for Student Success released a national report, dated October 2026, from a Spring 2026 survey of U.S. postsecondary students (202 completers; 188 in comparisons; 35% reported a disability or chronic health or mental health condition). Over 90% of both disabled and non-disabled students learned generative AI entirely on their own; instructor training and accommodations-office support were the least used sources. Students without disabilities reported equal or higher use across most areas. Most disabled students (82 to 92%) said AI tools met their access needs, yet only 6% reported AI training from instructors. Over 45% of both groups cited cost of paid versions as a barrier and over 40% cited course restrictions. All students worried about hallucinations and privacy; disabled students were more concerned about how others perceive their AI use (25% vs 18%).
Why it matters, evidence and limitations
- Why it matters
- Frames AI support as accessibility infrastructure: students, especially disabled students, find access value but get little institutional training or accommodations guidance, and cost and course policy limit use. Directly relevant to U.S. student-success and disability-services planning.
- Evidence and measured results
- Exploratory mixed-methods survey of a paid convenience panel; descriptive percentages with unadjusted exploratory comparisons.
- Limitations and uncertainty
- Convenience sample from a paid panel; not representative; exploratory unvalidated measure; small subgroups; no outcome data. Exact release day not stated.
How to read this edition
Source findings, measured results and limitations come from the cited publications. Patterns, operating questions, role takeaways and implementation considerations are Lighthouse Advisory interpretation, stated as questions to validate locally rather than guaranteed outcomes. Vendor and operator claims are labeled as claims. Full research method.
- Independent reporting
- Independent reporting with attributable sources but without a formal evaluation design.
- Vendor claim
- A supplier-provided assertion that has not been upgraded to independent evidence.
- Academic research
- Research produced through an academic institution or peer-reviewed venue.