1. Current problems shaping the undergraduate landscape
Undergraduate medical education (UME) is experiencing a fast pull from clinical practice toward artificial intelligence (AI) and digital medicine, but medical curricula often respond in a piecemeal way. Another noteworthy input comes from the Association of American Medical Colleges (AAMC), which asserts that medical education should include AI within its curriculum and learning process, while concurrently resolving the new challenges related to AI through collaborative efforts on an individual, institutional, and community level [
1]. Data from the Curriculum SCOPE snapshot revealed that the number of medical schools reporting the integration of AI within their curriculum has increased from 88 to 140 between 2023 and 2024, which represents a percentage increase from 53% to 77% [
2]. However, a snapshot from the same source revealed that approximately 52% of medical schools lacked an appropriate policy for AI use, and only 30% of the medical schools granted learners access to an AI platform.
At the same time, digital capability needs are broader than AI alone. The Digital Health Competencies in Medical Education (DECODE) international Delphi consensus (2025) describes a “dearth” of digital health education and proposes a structured set of competencies for medical graduates [
3]. This matters because many AI issues, including data governance, information systems, and patient-facing communication about technology, are inseparable from the wider digital health ecosystem described in the consensus.
A further problem is that popular generative AI tools can generate fluent but incorrect content, including fabricated references (“hallucinations”), which complicates both learning and scholarly integrity [
4]. In a recent scoping review on generative AI in medical education, recurring themes included concerns about academic integrity, data accuracy, and potential harms to learning [
5]. The authors also emphasize the need to develop learners’ skills in critical evaluation and to rethink assessment methods.
2. What recent research implies about “what to teach”
Current discussions increasingly converge on the idea that UME should teach competence for safe use, not tool-specific novelty. The DECODE framework defines digital health competencies across domains, including professionalism, information systems, health data science, and patient and population digital health, and explicitly positions it as a basis for systematic curriculum design [
3].
For AI specifically, a useful education-friendly structure is tiering. Schubert et al. [
6] describe a tripartite model for medical AI competence, which includes basic, which refers to the ability to use appropriate tools; proficient, which refers to critical appraisal of utility and outputs, as well as ethical considerations; and expert, which includes a higher level of understanding combined with expertise. Their framing is helpful for UME planning because it supports a realistic “minimum viable competence” consisting of basic and selected proficient elements, while reserving advanced technical depth for electives or later training.
Across both digital health and AI, current literature repeatedly points to critical appraisal, ethics, privacy, and communication as “cross-cutting” outcomes that can be integrated into clinical case discussions and early clinical-skills teaching rather than being isolated in a single informatics lecture [
2].
3. Solution directions most supported by the current discussion
A practical synthesis of current guidance identifies three solution directions that address the main failure modes: fragmentation, governance gaps, and weakness in assessment.
A first direction is explicit competency mapping, ideally connecting AI learning outcomes to an established digital health competency structure, such as DECODE, so that AI is taught as part of digital professionalism and data-informed care rather than as a standalone “extra” [
3].
A second direction is to treat governance and access as curriculum-enabling infrastructure. The AAMC snapshot suggests that policy and secure access lag behind curricular ambition, creating inequity, as some learners will use powerful tools while others cannot, and increasing unregulated and inconsistent use [
2]. The AAMC “Principles for the responsible use of AI in and for medical education” provide a concrete set of actionable principles, including human-centered focus, ethical and transparent use, equal access, privacy protection, monitoring and evaluation, and educator development [
1].
A third direction is assessment redesign that discourages “AI-as-shortcut” while rewarding interpretive and reflective skill. The AMEE Guide on AI and assessment emphasizes that educators and institutions are grappling with AI’s impact on assessment and highlights the importance of ethical issues, faculty development, and transparency (or acknowledgment) in assessment contexts [
7]. According to recent studies, if a clear ethical framework is in place, AI may improve unbiased evaluation in UME. One example is an AI-simulated patient tool designed to assess communication and reasoning. The AI technology offers immediate feedback on the basis of rubric criteria related to communication and decision-making skills. The importance of AI in this scenario lies in the necessity for reflection by the students on how well they did, as well as receiving feedback from AI technology. The ethical issues involved here include case reviews by experts and stringent patient data privacy. Pilot data reveal that students are happy with the AI and its instant feedback feature [
8].
Ethical framing must remain the foundation. The World Health Organization (WHO) provides ethical guidelines for the governance of AI, and the WHO recommendations for AI ethics and governance emphasize the importance of incorporating ethics and human rights into AI applications and utilization in the health sector, along with the associated risks and governance [
9].
4. Example model for a UME solution
A concise model that aligns with the above evidence is a spiral micro-curriculum embedded in preclinical systems blocks, designed to deliver “basic and selected proficient” competence while simultaneously shaping governance and assessment expectations.
One example structure is a four-session SAFE-AI spiral repeated across blocks (e.g., cardiology, endocrinology, and infectious diseases), with each session anchored in a short clinical vignette and requiring explicit disclosure of any AI use to normalize transparency rather than policing. This structure is consistent with three elements: (1) tiered competence thinking for safe use and critical appraisal [
6], (2) DECODE’s emphasis on professionalism, data, and information systems competence [
3], and (3) a governance-first approach consistent with AAMC principles and WHO ethics guidance [
1,
9].
“SAFE-AI” can be operationalized through four elements: Safety, including privacy red flags and source checking; Appraisal, including evidence, calibration and limits, and asking “what could be wrong?”; Fairness, including representativeness, bias, and inequity; and Explain, including patient-facing disclosure and shared decision-making about AI-supported care. This structure also directly addresses the documented risk that generative AI can fabricate references and misinformation [
4].
To implement a stage-specific spiral micro-curriculum for AI in UME education, one should aim to incorporate short and repeatable learning units into the existing course structure (
Fig. 1). During the initial pre-clinical period, the micro-curriculum will cover the topic of AI literacy and ethics based on short assignments (e.g., detecting hallucinations produced by the software, verifying the accuracy of the output information, and critical evaluation) during lectures and small group seminars. Similarly, in the latter pre-clinical period, the core AI competencies will be covered again, but at a higher level through critical appraisal with micro-units integrated into case-based learning in comparison of the reasoning offered by AI to clinical recommendations and analysis of bias and fairness. Finally, in the clinical setting, these core skills will be further developed in relation to practice, through a micro-curriculum covering topics related to decision-making based on AI output, communicating with the patient about the role of the algorithm in treatment, and professional accountability. In each phase of medical training, the same core skills, including safety, appraising AI output, fairness, and explaining to the patient, will recur with increased complexity.
Acknowledgements
None.
Funding
The author declares that no funds, grants, or other financial support were received for the conduct of this study.
Conflicts of interest
No potential conflict of interest relevant to this article was reported.
Author contributions
PD contributed to all aspects of the study, including conceptualization and design, data analysis and interpretation, manuscript drafting and revision, and approval of the final version of the manuscript.
Fig. 1.Spiral stage-specific artificial intelligence (AI) micro-curriculum embedded across undergraduate medical education. TBL: Team-based learning, PBL: Problem-based learning, OSCE: Objective structured clinical examination.
References
- 1. Association of American Medical Colleges. Principles for the responsible use of artificial intelligence in and for medical education. https://www.aamc.org/about-us/mission-areas/medical-education/principles-ai-use. Published January 3, 2025. Accessed March 13, 2026
- 2. Farmakidis AL, Singh A, Leaf K, Rossi A. Artificial intelligence curricula in U.S. and Canadian medical schools. Washington DC, USA: Association of American Medical Colleges; 2025.
- 3. Car J, Ong QC, Erlikh Fox T, et al. The digital health competencies in medical education framework: an international consensus statement based on a Delphi study. JAMA Netw Open. 2025;8(1):e2453131. https://doi.org/10.1001/jamanetworkopen.2024.53131
- 4. Eysenbach G. The role of ChatGPT, generative language models, and artificial intelligence in medical education: a conversation with ChatGPT and a call for papers. JMIR Med Educ. 2023;9:e46885. https://doi.org/10.2196/46885
- 5. Preiksaitis C, Rose C. Opportunities, challenges, and future directions of generative artificial intelligence in medical education: scoping review. JMIR Med Educ. 2023;9:e48785. https://doi.org/10.2196/48785
- 6. Schubert T, Oosterlinck T, Stevens RD, Maxwell PH, van der Schaar M. AI education for clinicians. EClinicalMedicine. 2024;79:102968. https://doi.org/10.1016/j.eclinm.2024.102968
- 7. Masters K, MacNeil H, Benjamin J, et al. Artificial intelligence in health professions education assessment: AMEE Guide No. 178. Med Teach. 2025;47(9):1410-1424. https://doi.org/10.1080/0142159X.2024.2445037
- 8. Thesen T, Alilonu NA, Stone S. AI patient actor: an open-access generative-AI app for communication training in health professions. Med Sci Educ. 2024;35(1):25-27. https://doi.org/10.1007/s40670-024-02250-2
- 9. World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance. Geneva, Switzerland: World Health Organization; 2021.
Citations
Citations to this article as recorded by
