UNESCO's AI Ethics Framework: From Clinical Safety to Systemic Benefits and Risks for Health
Nikolay Lipskiy, MD, PhD, MBA
CAMA CEO and Principal scientist
An AI application can perform well while the healthcare system around it performs worse.
That possibility should be central to how we evaluate artificial intelligence (AI) in healthcare. Assessing individual applications is essential, but it does not, by itself, establish whether their integration improves care delivery, population health, equity, or health-system resilience.
UNESCO’s Recommendation on the Ethics of Artificial Intelligence (2021) offers a valuable foundation for this broader perspective. Its implementation is supported by two complementary instruments: the Ethical Impact Assessment (EIA) and the Readiness Assessment Methodology (RAM), including RAM 2.0, introduced in September 2026.
Connecting Health and Social Well-being
One particularly important feature of UNESCO’s Recommendation is Policy Area 11: Health and Social Well-being (paragraphs 121–130).
Rather than limiting attention to the technical performance of clinical AI applications, this policy area addresses a broader range of considerations, including:
Safety, effectiveness, and scientific and medical validation of AI applications.
Human oversight, informed consent, and patient participation.
Privacy, discrimination, and protection of people in vulnerable situations.
Mental health, human relationships, and psychosocial well-being.
Importantly, paragraph 129 explicitly encourages research into the effects of AI technologies on health-system performance and health outcomes. This creates a direct connection between the evaluation of individual AI applications and their broader consequences for healthcare.
Building on this perspective, I propose distinguishing between systemic AI benefits and systemic AI risks for health.
This distinction also creates an opportunity to consider AI not only as a clinical technology but as a systemic factor influencing human health and well-being across the broader sociotechnical environment.
What Makes AI Benefits and Risks Systemic?
Systemic AI benefits are not simply the sum of improvements achieved by individual applications.
They may emerge when AI integration improves coordination across healthcare services, strengthens public health decision-making, supports continuity of care, reduces inequities, and increases health-system resilience.
For example, AI-enabled integration of clinical, laboratory, and population-health information may improve the coordination of prevention and treatment across multiple healthcare organizations. The resulting benefits depend not only on individual AI applications but also on how effectively these applications interact within the broader healthcare system.
Similarly, systemic AI risks are not merely the combined risks of individual AI systems.
They may emerge through interactions, dependencies, feedback loops, and shared vulnerabilities across technologies, clinical workflows, healthcare organizations, information infrastructures, governance arrangements, and populations.
Consider a hypothetical example: AI-supported triage accelerates initial patient assessment but increases referrals into already constrained specialist services. If adopted widely without coordinated capacity planning, a local efficiency gain could contribute to longer downstream waiting times and unequal access to care.
What makes such risks systemic is their potential to propagate, amplify, or generate consequences across interconnected components, sometimes affecting the performance and resilience of the healthcare system as a whole.
Importantly, systemic benefits and risks may coexist. Improved efficiency in one service does not necessarily produce better overall outcomes and may introduce new vulnerabilities elsewhere.
These are proposed analytical concepts, not formal benefit or risk categories established by UNESCO’s Recommendation.
From Ethical Principles to AI Risk Governance
UNESCO’s implementation instruments offer complementary perspectives on responsible AI development and deployment.
Ethical Impact Assessment (EIA) supports assessment of AI systems throughout their lifecycle. It examines alignment with ethical principles, identifies potential positive and negative impacts, and supports stakeholder engagement and measures to address unintended consequences.
Readiness Assessment Methodology (RAM 2.0) evaluates countries’ preparedness to develop, deploy, and govern AI responsibly. It identifies strengths and gaps in institutional capacities, policies, regulatory frameworks, and technical infrastructure.
Importantly, RAM 2.0 includes a dedicated Policy Area 11: Health and Social Well-being, examining national healthcare AI policies, regulatory instruments, risk-management requirements, clinical safety and effectiveness, human oversight, incident reporting, and monitoring of mental-health and psychosocial impacts.
Together, these instruments connect the assessment of AI-system impacts with the broader institutional conditions needed for responsible AI governance.
They provide a foundation for developing more explicit approaches to systemic healthcare AI assessment, although neither instrument constitutes a complete methodology for evaluating systemic healthcare AI benefits and risks.
The Next Challenge for Healthcare AI Governance
Assessments of individual AI systems remain essential, including their technical performance, clinical safety, effectiveness, and ethical impacts.
However, healthcare organizations and public health authorities must also evaluate how AI changes relationships among services, organizations, infrastructures, and populations.
A systems perspective raises several important questions:
Does AI integration improve clinical outcomes and population health?
Does it strengthen coordination, continuity of care, and health-system resilience?
How does it redistribute workload, resources, and access to healthcare?
Does it introduce shared dependencies, cascading failures, or new vulnerabilities?
Which populations receive the benefits, and which bear the risks?
Can governance arrangements detect and respond to changing effects over time?
Answering these questions requires moving beyond isolated evaluations of AI applications toward system-level assessment, continuous monitoring, and governance of AI-enabled healthcare.
The central question is not only whether an AI system is safe and effective. It is also whether integrating AI makes healthcare systems—and the populations they serve—healthier, safer, more equitable, and more resilient.
UNESCO’s explicit connection between health, social well-being, and health-system performance provides an important ethical foundation for recognizing AI as a systemic factor influencing human health across the broader sociotechnical environment.
The next challenge is to translate this perspective into explicit methods for identifying, evaluating, and governing systemic AI benefits and risks—not only within healthcare organizations but also across the wider societal systems that influence population health and well-being.
UNESCO References
UNESCO (2021). Recommendation on the Ethics of Artificial Intelligence. Particularly Policy Area 11: Health and Social Well-being, paragraphs 121–130.
UNESCO (2026). Readiness Assessment Methodology (RAM 2.0). Particularly Policy Area 11: Health and Social Well-being.
UNESCO. Ethical Impact Assessment (EIA). Implementation instrument supporting assessment of ethical impacts associated with AI systems.
Comments