Answers are grounded in this paper and cite their sources.
Agentic artificial intelligence is entering higher education as systems progress from generating answers to planning and executing actions. This transition creates an ethical problem that aggregate accuracy cannot resolve: institutions may delegate consequential discretion before they possess evidence of durable benefit, fairness, or effective human control. This critical integrative review examines 15 peer-reviewed reviews, one recent preprint retained separately, and selected primary studies to determine what the evidence supports and what remains ethically unjustified. The literature is conceptually unstable, often conflating perceived agency, static generative AI, and systems with goal-directed execution. Evidence is strongest for bounded artifact quality and weakest for durable learning, equity, workload, and institution-level outcomes. Authentic deployments remain mostly short, single-site, and weakly connected to tested oversight arrangements. The article develops the Autonomy-Oversight-Evidence framework, comprising five autonomy levels, five oversight levels, and six stages of evidence maturity. Its normative principle is that delegated autonomy must not exceed either the maturity of the supporting evidence or the strength of accountable human oversight. This principle follows from duties to prevent avoidable harm, preserve learner autonomy, distribute risks fairly, maintain contestability, and keep institutional accountability with responsible persons. The framework converts these duties into reporting requirements, permission boundaries, approval checkpoints, logging, escalation, appeal, and rollback. Agentic AI may improve selected educational processes, but current evidence does not justify broad claims of sustained effectiveness or autonomous use in rights-affecting decisions.
No takes yet. Share an insight, caveat, or question.
Abhishek Dey (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: