If your view of medical AI ethics in 2025 assumes the research community is broadly addressing the full range of ethical concerns (algorithm accountability, governance, biosecurity, global health equity) proportional to their importance, the bibliometric analysis of PubMed Central from 2021-2025 shows the actual research landscape is concentrated. Medical AI ethics publications more than doubled in 2025, reaching 2,378 papers (up from 1,114 in 2024). Within this growth, governance topics accounted for 1,228 publications (52% of ethics-tagged publications). Algorithmic concerns: 896 (38%). Societal concerns: 874 (37%). But biosecurity, the area of greatest concentrated policy attention and arguably the highest-stakes ethical concern, was discussed in only 14 publications in 2025. Global health equity was similarly underexplored relative to its importance. The medical AI ethics conversation is expanding in volume but remains narrow in scope.

The bibliometric analysis methodology is direct: PubMed Central publications from January 2021 to December 2025, identified using search terms for medical AI and ethics, then categorised by emphasis on data sharing, algorithm sharing, biosecurity, and global health. Ethics topics were grouped under algorithmic, governance, or societal concerns. The doubling of publications between 2024 and 2025 (1,114 to 2,378) is the topline data observation; the topic concentration is the contested observation.

The topic distribution by year:

  • Governance (institutional policies, regulatory frameworks, oversight structures): 130 (2021) → 1,228 (2025). Largest category. Driven by EU AI Act activity, FDA guidance development, healthcare-specific governance frameworks.
  • Algorithmic (algorithm accountability, fairness, bias, explainability): 107 → 896. Second-largest category. Driven by accumulating evidence on algorithmic bias in clinical AI.
  • Societal (equity, justice, accessibility, broader social impact): 72 → 874. Substantial growth but smallest of the three primary categories.
The data observation: governance is the dominant topic, and the dominance is growing. The 2025 governance share (52%) is higher than the 2024 share (49%) and substantially higher than the 2021 share (39%). The research community is increasingly oriented toward governance questions and proportionally less oriented toward algorithmic and societal questions.

The biosecurity gap is the most striking finding. Despite the attention paid to biosecurity in policy discussions, the subject is relatively unexplored in medical AI publications: in 2025, only 14 of these publications discussed biosecurity, with even fewer directly addressing the ethical implications of misuse or dual use. Biosecurity concerns about medical AI (particularly around protein design AI that could enable biological weapon development, drug discovery AI that could be repurposed, and clinical AI that could be deployed to harm) are arguably among the highest-stakes ethical concerns in the field. The research community has not engaged with them at scale.

Three structural drivers explain the governance dominance and the biosecurity gap.

The first driver: governance research is operationally tractable. Researchers can write meaningful papers on governance frameworks, regulatory approaches, institutional policies, and oversight structures using publicly available information (regulations, policies, organisational documents). The research methodology is accessible to researchers without specialised biological or technical capability. The publication pathway is established.

The second driver: biosecurity research is structurally constrained. Researching biosecurity risks in medical AI requires either specialised biological capability (to assess what protein design AI or drug discovery AI could enable), specialised AI security capability (to assess what model misuse could produce), or classified-information access (to engage with state and non-state biosecurity threats). The research community capable of conducting this work is small. The publication pathway is constrained: some biosecurity research cannot be openly published because publishing it would aid misuse. The dual-use research dilemma applies.

The third driver: societal and equity research is under-funded and under-recognised compared to governance and algorithmic research. The 874 societal publications in 2025 include important work on equity, justice, and accessibility, but the research base is structurally smaller than governance and algorithmic research. The funding flows toward governance (where regulation creates demand for research input) and algorithmic (where industry demand for fairness/bias work exists). Societal research depends more on academic and public funding, which has expanded but not at the same rate.

The contested observation: the medical AI ethics research landscape in 2025 is structurally over-weighted toward governance and structurally under-weighted toward biosecurity, global health equity, and broader societal impact. The research output is not proportional to the underlying ethical concerns.

The global health exception is informative. Publications addressing global health AI (193 total in 2025) showed a different ethics topic distribution. Societal concerns (54 publications) led the global health AI ethics literature, surpassing both governance (43) and algorithmic concerns (29). In the global health context, equity, justice, and accessibility are the primary ethical considerations, and researchers working in this space write about them at higher proportion than the broader medical AI ethics literature.

The geographic distribution of global health AI ethics research is also informative. Europe leads (38 publications in 2025), followed by East Asia (31), North America (28), South Asia (14), and Middle East (11). Sub-Saharan Africa (8), Southeast Asia (5), North Africa (4), Latin America (3), and Oceania (3) are substantially underrepresented relative to the populations these regions serve. The research about AI for global health is being done predominantly outside the regions where global health AI deployment matters most.

Three implications follow for organisations operating in or adjacent to medical AI ethics in 2026.

The first implication (contested): the medical AI ethics research landscape needs proportional rebalancing toward biosecurity. The 14 biosecurity publications in 2025 reflect a research base inadequate to the scale of biosecurity concerns about medical AI. Research funders, institutions, and individual researchers with capability to engage with biosecurity should treat this as a high-priority gap. The path forward is contested: opening biosecurity research risks aiding misuse, but the alternative (continued under-engagement) leaves the field without the ethical reasoning infrastructure for emerging biosecurity-relevant AI capabilities.

The second implication (contested): governance research has reached or is approaching diminishing returns relative to other ethics topics. The 52% governance share, growing from 39% in 2021, suggests the research community is converging on governance topics in a way that may be over-investing relative to algorithmic, societal, and biosecurity concerns. Some rebalancing of research attention may produce more proportional ethics coverage of the actual landscape of medical AI ethics concerns.

The third implication (contested): the global health AI ethics research base requires structural geographic rebalancing. The current distribution (heavy in Europe, East Asia, North America; thin in sub-Saharan Africa, Latin America, Southeast Asia) produces research output that may not engage adequately with the contexts where global health AI deployment matters most. Research investment in underrepresented regions, and partnership models that centre local researchers in research design and authorship, are operational responses to this gap.

The contested observation: medical AI ethics research is expanding rapidly in volume but the topic distribution and geographic distribution show structural gaps that the current research community has not closed. The strategic question for 2026-2028: whether the research community can develop the infrastructure (funding, capability, methodological frameworks, publication pathways) to engage with biosecurity, global health equity, and underrepresented geographies at scale. The answer will shape what medical AI ethics looks like through the 2030s.

For organisations setting medical AI policy, deployment, or research strategy in 2026, the contested observation calibrates expectations: the existing medical AI ethics research base is substantial on governance topics, weaker on algorithmic and societal topics, and structurally inadequate on biosecurity and underrepresented geographies. Strategic plans that operate from this awareness (engaging with the strong research where it exists, commissioning research where it's weak, and being cautious about claiming ethics coverage in areas where research is thin) will produce better aligned outcomes than plans that treat the ethics research base as uniformly developed.

The methodological note: the 14-publication biosecurity count is a single year and may understate the actual research occurring in classified or industry settings. The publicly available research base is what the analysis captures. The actual capability for ethical reasoning about biosecurity-relevant medical AI may be somewhat better than the bibliometric data suggests, but the publicly available base is what most policy, regulatory, and institutional decision-making draws on. The publicly available base is the structural variable for governance and deployment decisions.