Global Medical AI Policies Score Well on Vision but Falter on Enforcement, Study Finds

Artificial intelligence is rapidly reshaping medicine, from diagnostic imaging algorithms that flag tumors earlier than the human eye to large language models that draft clinical notes and triage patients. Yet while the technology races ahead, the policies meant to govern it have often been judged by impression rather than evidence. A new study published in BMC Health Services Research changes that, offering one of the first quantitative, cross-national audits of how well governments are actually writing their medical AI rules. The verdict is striking: the world’s major economies are producing ambitious, well-framed strategic visions, but the operational machinery needed to turn those visions into enforceable clinical safeguards is lagging behind.

The research team, led by Fei Wang, Siqi Huang, Yiwen Xu, Jin Wen and Xing Qu, affiliated with Sichuan University’s West China Hospital and its Business School, analyzed forty-two national-level medical AI policy documents issued between 2019 and 2025 by seven governments and blocs: the United States, the United Kingdom, Japan, South Korea, Australia, China and the European Union. Rather than simply reading and summarizing the documents, the researchers applied the Policy Modeling Consistency, or PMC, Index model, a quantitative framework originally developed in policy science to measure how internally coherent, structurally complete and implementation-ready a piece of policy text really is.

The PMC methodology is technically intricate. The team first constructed a multidimensional evaluation system comprising nine primary variables, each subdivided into a total of thirty-five secondary variables. These dimensions capture everything from the nature and purpose of a policy to its content depth, the policy tools it deploys, the breadth of perspective it adopts, and the strength of its evaluation and feedback mechanisms. Each secondary variable is scored, the scores are combined into a composite PMC index, and a concavity index is calculated to grade the policy on a five-point scale from A to E. The final step is visualization: a so-called PMC surface map renders each policy’s strengths and weaknesses as a three-dimensional landscape, allowing researchers to see at a glance where a document bulges with detail and where it collapses into vagueness.

The headline numbers reveal a field that is competent but not exceptional. Of the forty-two policies analyzed, only two, or 4.76 percent, earned the top Grade A rating, while twenty-nine, or 69.05 percent, landed in Grade B. Nine policies, listed as 14.51 percent in the study, were rated Grade C, and two, or 4.76 percent, fell to Grade D. The overall mean PMC index came out at 6.67, which the authors interpret as above-average global policy quality. In other words, most governments are doing reasonably well at writing medical AI strategies, but almost none are producing documents the model considers excellent.

More revealing than the aggregate scores is the pattern beneath them. The policies scored highest on dimensions such as policy nature, policy evaluation and policy perspective, indicating that governments excel at declaring broad intent, articulating why medical AI matters, and framing their strategies within national and global contexts. The weakest scores, by contrast, appeared in policy content and policy tools, the very dimensions that determine whether a strategy can actually be executed. The authors describe this as a structural imbalance: comprehensive strategic frameworks paired with weak implementation instruments. A government can proclaim that AI will transform its health system, but if the document lacks concrete regulatory levers, funding mechanisms, accountability structures and measurable milestones, the proclamation remains aspirational prose.

This imbalance has real consequences for patient safety and innovation alike. Medical AI systems touch some of the most sensitive territory in modern governance: clinical decision-making, patient data privacy, algorithmic bias, liability when an algorithm errs, and the ethical boundaries of machine involvement in care. The study found that most of the policies it examined exhibit robust strategic orientation but insufficient regulatory and ethical detailing. That gap matters because the absence of specific rules does not create a regulatory vacuum in practice; it simply shifts decisions to courts, hospitals and individual clinicians, often after harm has occurred or trust has eroded.

The seven jurisdictions studied represent the full spectrum of approaches to AI governance. The United States has leaned on agency-level guidance, including the Food and Drug Administration’s evolving framework for AI-enabled medical devices, alongside broader executive initiatives. The United Kingdom has pursued a principles-based, regulator-led model through bodies such as the Medicines and Healthcare products Regulatory Agency and the Department for Science, Innovation and Technology. The European Union has embedded health AI within its sweeping horizontal AI legislation and its General Data Protection Regulation, prioritizing fundamental rights. China, Japan and South Korea have issued national strategies that couple industrial policy with health-sector pilot programs, while Australia has taken a more incremental, standards-driven path. The PMC analysis allows these philosophies to be compared on a common quantitative footing for the first time at this scale.

For China in particular, the study offers pointed recommendations. The authors conclude that the country’s policy framework should prioritize refining data governance mechanisms, strengthening ethical accountability, and building dynamic policy adjustment systems capable of keeping pace with fast-moving technology. The emphasis on dynamism is notable: static policy documents written for the AI of 2020 may be poorly matched to generative models and foundation-model medical applications emerging today. A policy system that cannot revise itself quickly risks perpetually regulating the previous generation of technology, a problem the authors suggest is not unique to China but visible across the dataset.

The broader lesson for policymakers everywhere is that ambition and operability must be designed together. The PMC Index model, with its decomposition into nine primary and thirty-five secondary variables, effectively functions as a diagnostic checklist for policy drafting. Documents that score well tend to specify who is responsible for implementation, how compliance will be monitored, what resources are committed, and how success will be measured. Documents that score poorly tend to be heavy on vision statements and light on instruments. As governments worldwide prepare the next wave of AI health regulations, the study’s methodology offers a template for stress-testing drafts before they are published, rather than discovering their weaknesses through implementation failures.

The research also demonstrates the value of treating policy text itself as data. By combining text mining with quantitative modeling, the team converted hundreds of pages of governmental prose into comparable scores and visual surfaces, making cross-national policy assessment reproducible rather than subjective. In a domain where hype routinely outpaces evidence, that kind of rigor is itself a form of governance. The study, published open access on 5 October 2026 with the DOI 10.1186/s12913-026-15763-w and supported by the National Natural Science Foundation of China, suggests that the world’s governments know where they want medical AI to go. What remains to be seen, and what their current policies too rarely specify, is exactly how they plan to get there.

Subject of Research: Quantitative evaluation of national-level medical artificial intelligence policies using the Policy Modeling Consistency Index model

Article Title: Evaluation of global national-level medical artificial intelligence policies and strategies based on the policy modeling consistency index model

Article References: Wang, F., Huang, S., Xu, Y., Wen, J., & Qu, X. (2026). Evaluation of global national-level medical artificial intelligence policies and strategies based on the policy modeling consistency index model. BMC Health Services Research. https://doi.org/10.1186/s12913-026-15763-w

Image Credits: AI Generated

DOI: 10.1186/s12913-026-15763-w

Keywords: medical artificial intelligence, health policy, PMC index model, policy evaluation, policy implementation, data governance, AI regulation, health services research, comparative public policy, clinical AI, policy quality assessment, China