AI Is Changing Medicine at Its Core. Is Medical Education Keeping Up?

Healthcare has a reputation for being slow to adopt new technology. Yet in a recent survey, nearly a third of UK physicians said they use AI in their work every week, and 66% said they have no access to any training in it (Royal College of Physicians (RCP), 2026).¹ On its own, that might not sound alarming. But the NHS has already received its first negligence claims in which AI may have played a role (Clover and Devereux, 2026, cited in Sokol, 2026).² ³

In most situations, a wrong answer from a chatbot is an inconvenience. In a medical consultation, it can mean a missed diagnosis, or worse.

The government has not stayed on the sidelines as AI changes how healthcare works and what future doctors need to learn. In July 2025, the 10 Year Health Plan for England promised to update what doctors are taught within three years, including training in AI (Department of Health and Social Care, 2025).⁴ Even after the recent change of prime minister, that promise remains official policy. But it is still an open question whether this plan will actually be followed, or whether it is only a direction, with concrete steps to come later.

Current trends already show a growing market of AI training courses, including courses for doctors. But truly structured, in-depth programmes can be counted on one hand, especially in the UK. Most of these are year-long fellowships, which take only a few dozen doctors a year and are tied to NHS employment, leaving doctors who work outside the NHS largely shut out. Consequently, preparing the whole workforce remains an aspiration.

The remaining options are short courses: one-day, one-week or online. That is still real progress, but there is little evidence that training this brief gives doctors the depth they need to work with AI safely (Woods et al., 2026).⁵ In other words, AI is changing medicine at its core, while medical education is changing only at its edges.

Practice is moving faster than training

To understand this better, it helps to look at how medical schools teach AI. A 2023 survey found that most UK medical schools offered no dedicated AI teaching. Those that did offered very little: only a few hours across the five-year course (Health Data Research UK and Medical Schools Council, 2023).⁶ This is not structured learning, and there is still no national standard for what AI teaching should include. In part, this is understandable. Generative AI tools such as ChatGPT only became widely available at the end of 2022, and they have changed rapidly every year since.

Despite the lack of structured education, AI is spreading quickly across the profession. Among UK GPs (family doctors) who took part in national surveys, the use of generative AI tools, most often ChatGPT, rose from 20% in 2024 to 52% in early 2026 (Blease et al., 2026).⁷ These surveys rely on doctors’ own reports and may have attracted the doctors most interested in AI, so the figures may be somewhat biased. Even so, they show a clear upward trend. Many apply it to clinical tasks, such as suggesting a diagnosis or treatment options. Yet in 2025, 95% of GPs said they had received no professional training in these tools, which left them to learn how to navigate AI on their own, and 85% said their employer had never encouraged it (Blease et al., 2025).⁸ As the survey authors warn, AI risks becoming normalised in primary care “by default rather than by design” (Blease et al., 2026).⁷

It is also important not to treat ChatGPT and certified medical AI tools as the same thing. Doctors need to know which kind of tool they are using, how far to trust its output, and what to do when a result looks wrong. General-purpose tools such as ChatGPT were not designed for medical use. This affects the doctor’s responsibility, as the doctor carries the full weight of checking every answer. By contrast, with a certified medical AI tool, the doctor must know what it was approved for and use it only for that. For example, a tool designed to detect diabetic eye disease is not designed to spot other eye conditions (Sokol, 2026).³

Good intentions, but little evidence

There are already early shifts that are likely to set off a wider transformation. In 2025, the body that represents UK medical schools (the Medical Schools Council), together with Health Data Research UK, formally recommended that data science and AI be included in what medical students are taught (Medical Schools Council and Health Data Research UK, 2025).⁹ The Royal College of Physicians has gone further, calling for digital and AI skills to be built in at all career stages (RCP, 2026).¹

Still, one problem remains: AI tools are being adopted fast, but the research on how to teach them has not caught up. There is no shortage of ideas about what doctors should learn. A 2026 review of 54 studies from 22 countries brought together many such proposals. But most of these sources were editorials and opinion pieces, not research backed by evidence (Hunt et al., 2026).¹⁰

Would doctors accept it?

The obvious question is how welcoming medical students and doctors will be to adding AI training to their already busy schedules. The evidence suggests they are more open to it than one might expect. A review of 26 studies from around the world found that 83% of medical students believe they need AI training, yet only around a third feel confident using AI in clinical practice (Mastour et al., 2026).¹¹

Qualified doctors want it too, mainly as protection. A Nuffield Trust and RCGP study reported that 89% of GPs who don’t use AI, and 80% of those who do, were not opposed to the technology. Their main concern was who is legally responsible if something goes wrong (Kumpunen et al., 2025).¹² Even NHS Resolution, which handles negligence claims against the NHS, says staff need training not only in how to use AI tools, but in their limitations and risks (NHS Resolution, 2026).¹³

Making room for AI competence

Beyond the speed of change, there is a more practical barrier: medical programmes are already overloaded. In the same review of 26 studies, overcrowded programmes were the most common barrier to teaching AI, reported in 68% of the studies that looked at barriers (Mastour et al., 2026).¹¹ At the same time, the skills doctors need are expected to change. More than half of UK GPs believe AI will affect how much doctors need to memorise, such as drug doses, drug interactions, normal ranges for test results or the details of clinical guidelines (Blease et al., 2026).⁷ When AI can find these facts in seconds, doctors need a specific kind of clinical judgement and a feel for these technologies. This cannot be developed in a short course: it has to be built up gradually, with regular assessment along the way. Such judgement means recognising when AI gives a convincing answer that is wrong, how certain its answer really is, when it may work less well for patients from certain demographic groups, when automation bias, the tendency to trust a machine over one’s own judgement, may take over, and when its recommendations must be challenged (Lewis et al., 2025).¹⁴ It also means knowing whether a tool has been clinically validated and approved for use in the UK.

But relying on AI too heavily carries two kinds of risk. The short-term risk is that a doctor who trusts AI too much may follow its mistakes without checking them, which could put patients at risk. The long-term risk is that doctors who routinely hand their thinking over to AI could slowly lose the clinical reasoning and diagnostic skills they need to catch its mistakes. As Lewis et al. (2025)¹⁴ point out, even in the era of autopilot, pilots still practise flying manually. Doctors need to do the same.

It is therefore reasonable to suggest that a full restructure of the medical curriculum may not be necessary. A more logical step is to build AI training into existing subjects throughout medical school, starting with the basics and growing in depth as students reach their clinical years. For example, when students learn how to communicate with patients, a short mandatory introduction to explaining AI use could be added. When they study data ethics, a separate session could cover how AI fits into that area. In this way, AI would be added gradually to existing courses, rather than taught as a separate general block. This also matters because AI is used differently across specialties: its role in gynaecology, surgery, internal medicine or urology will not be the same. Integrating AI into existing modules, with a focus on how it applies to each specialty, is one of the approaches the review authors suggest (Mastour et al., 2026).¹¹

As with any part of medical training, AI skills should be assessed both in practical clinical situations and in written exams. And the teachers who train students in these technologies need training themselves: in the review by Mastour et al. (2026),¹¹ a lack of AI expertise among teaching staff was one of the most common barriers reported. Even then, the evidence that AI training improves outcomes for patients is still very limited (Woods et al., 2026).⁵

Doctors are caught between two risks

They are already using AI without any formal training, and the first negligence claims in which AI may have played a role have already arrived, although it is worth acknowledging that these cases do not yet show that AI caused harm. At the same time, some legal experts already suggest that not using AI may itself become grounds for a negligence claim in the future (Sokol, 2026).³

The solution is likely to be mandatory AI training, and it has to include assessment. The MLA, the national licensing exam that every new doctor in the UK must pass, does not currently mention AI anywhere in the list of topics it covers (General Medical Council (GMC), 2026a).¹⁵ Training that is never assessed is easy to treat as optional. So, the approach has to be comprehensive, including changes to what doctors are examined on. The GMC is currently reviewing its education framework, which makes this the right moment to do it (GMC, 2026b).¹⁶

Three further changes would help. First, for doctors already in practice, AI skills should be built into the annual reviews and five-yearly checks they already go through to keep their licence, not delivered as a click-through module that ticks a box.

Second, access cannot depend on who employs you. Locum (temporary) GPs are among the heaviest users of AI for diagnosis (Blease et al., 2026),⁷ yet they, along with doctors in the independent sector and many international medical graduates, often struggle to access structured training. Elsewhere, this is already a legal duty: since February 2025, the EU AI Act has required organisations using AI to take measures to ensure their staff have sufficient AI literacy (European Parliament and Council of the European Union, 2024).¹⁷

Third, the research on how to teach AI needs proper funding, so that training is built on evidence rather than opinion.

Training alone, however, cannot guarantee that AI is safe. Safe AI also depends on good governance and clear regulation: rules for introducing new tools, the certification and studies tools must undergo to show they are safe in practice, ongoing monitoring once they are in use, and clarity about who is responsible when something goes wrong.

Conclusion

AI is already part of medicine. If medical education does not respond seriously to the changes already happening in clinical practice, it risks being forced to change anyway: not by choice, but through the courts.

References

1. Royal College of Physicians (2026) Snapshot of UK physicians: artificial intelligence in healthcare. London: RCP. Available at: https://www.rcp.ac.uk/policy-and-campaigns/policy-documents/snapshot-of-uk-physicians-artificial-intelligence-in-healthcare/ (Accessed: 16 October 2026).
2. Clover, B. and Devereux, E. (2026) ‘Exclusive: first AI clinical negligence claims lodged’, Health Service Journal, 11 September. Available at: https://www.hsj.co.uk/patient-safety/exclusive-first-ai-clinical-negligence-claims-lodged/8124278.article (Accessed: 16 October 2026).
3. Sokol, D. (2026) ‘Six rules for doctors using AI’, BMJ, 394, e100883. doi: 10.1136/bmj-2026-100883.
4. Department of Health and Social Care (2025) Fit for the future: 10 Year Health Plan for England. London: DHSC. Available at: https://www.gov.uk/government/publications/10-year-health-plan-for-england-fit-for-the-future (Accessed: 16 October 2026).
5. Woods, L., Lyons, K., Van Der Vegt, A., Olsen, Q., Huang, W., Khor, J.S., Xu, N. and Sullivan, C. (2026) ‘Assessing the effectiveness of artificial intelligence education and training for healthcare workers: a systematic review’, BMC Medical Education, 26, 549. doi: 10.1186/s12909-026-08969-3.
6. Health Data Research UK and Medical Schools Council (2023) Survey of data science in UK medical school curricula. Available at: https://www.hdruk.ac.uk/wp-content/uploads/2023/11/Survey-of-Data-Science-in-UK-Medical-School-Curricula-Report-August-2023.pdf (Accessed: 16 October 2026).
7. Blease, C., Garcia Sanchez, C., Gaab, J., Alderman, J.E., Allen, L. and Kumpunen, S. (2026) ‘Generative artificial intelligence adoption and the future of the primary care profession: an online survey of UK general practitioners’ experiences and opinions’, BMJ Digital Health & AI, 2(1), e000253. doi: 10.1136/bmjdh-2026-000253.
8. Blease, C., Hagström, J., Garcia Sanchez, C., Kharko, A., McMillan, B., Gaab, J., Brulin, E., Locher, C., Hägglund, M., Riggare, S. and Mandl, K.D. (2025) ‘General practitioners’ adoption of generative artificial intelligence in clinical practice in the UK: an updated online survey’, Digital Health, 11. doi: 10.1177/20552076251394287.
9. Medical Schools Council and Health Data Research UK (2025) Data science in the medical curriculum: equipping medical students for the digital age. Available at: https://www.medschools.ac.uk/wp-content/uploads/2025/04/Data-Science-in-the-Medical-Curriculum-.pdf (Accessed: 16 October 2026).
10. Hunt, V.M. et al. (2026) ‘What the AI era doctor should know: a scoping review of proposed artificial intelligence competencies for medical education’, npj Digital Medicine. doi: 10.1038/s41746-026-02761-9.
11. Mastour, H., Mirzaei, S., Sohrabi, S. and Toofaninejad, E. (2026) ‘Is it time to introduce an artificial intelligence curriculum in undergraduate medical education? Medical students’ perspectives: a systematic review and meta-analysis’, BMC Medical Education, 26, 1380. doi: 10.1186/s12909-026-09630-9.
12. Kumpunen, S., Lobont, C., Garrard, L., Fisher, D., Lau, R., Fallica, G. and Fisher, R. (2025) How are GPs using AI? Insights from the front line. London: Nuffield Trust. Available at: https://www.nuffieldtrust.org.uk/research/how-are-gps-using-ai-insights-from-the-front-line (Accessed: 16 October 2026).
13. NHS Resolution (2026) NHS Resolution guidance on scheme coverage and liability issues concerning the use of Artificial Intelligence (AI). London: NHS Resolution. Available at: https://resolution.nhs.uk/?p=20686 (Accessed: 16 October 2026).
14. Lewis, M., Fraile Navarro, D., Blease, C., Shah, R., Riggare, S., Delacroix, S. and Lehman, R. (2025) ‘Clinical competencies for using generative AI in patient care’, BMJ, 391, e085324. doi: 10.1136/bmj-2025-085324.
15. General Medical Council (2026a) MLA content map. London: GMC. Available at: https://www.gmc-uk.org/cdn/documents/20260820—mla-content-map–english–september-2026-update_pdf-115633825.pdf (Accessed: 16 October 2026).
16. General Medical Council (2026b) Business plan 2026. London: GMC. Available at: https://www.gmc-uk.org/cdn/documents/business-plan-2026-english_pdf-113557173.pdf (Accessed: 16 October 2026).
17. European Parliament and Council of the European Union (2024) ‘Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act)’, Official Journal of the European Union, L, 12 July.