Medical AI · Regulation, ethics & governance · Lesson 6 of 7

WHO guidance on AI ethics and large multimodal models

Reviewed by Dr. Ram Kumar G, MBBS, MD (Internal Medicine), MPH (USA), on 2026-10-18 · Last updated 2026-10-18

Beginner
Clinical focus
About 25 minutes
Includes exercise

Before this lesson: The healthcare-AI landscape: what is in use, what works, and what to ask

In short

The World Health Organization's 2021 guidance set out six principles for ethical AI in health: protect autonomy; promote well-being, safety and the public interest; ensure transparency and explainability; foster responsibility and accountability; ensure inclusiveness and equity; and promote AI that is responsive and sustainable. Its 2024 guidance on large multimodal models applies these to generative AI, with recommendations for governments, developers and health providers. WHO has also issued regulatory considerations (2023) and, in 2026, recommendations on ethics review of AI-related health research.

You will learn to

List and explain WHO's six ethical principles for AI in health
Summarise WHO's guidance on large multimodal models and who it is aimed at
Apply the principles to a concrete AI deployment decision
Identify where ethics guidance and law reinforce each other

Six principles

WHO, Ethics and governance of artificial intelligence for health, 2021.
Principle	What it asks in practice
Protect human autonomy	Humans stay in control of health decisions; patients' consent and privacy are respected
Promote human well-being, safety and the public interest	Tools meet safety, accuracy and effectiveness requirements for their intended use
Ensure transparency, explainability and intelligibility	Enough information is published and understandable to the people affected
Foster responsibility and accountability	It is clear who is responsible, and there are ways to question decisions and seek redress
Ensure inclusiveness and equity	Tools are designed for and work across the populations that use them
Promote AI that is responsive and sustainable	Tools are evaluated in real use and their environmental and workforce effects are considered

Large multimodal models

WHO's January 2024 guidance on large multimodal models (LMMs, which include large language models) describes uses from diagnosis and patient-facing chat to administration, research and education, and the risks: inaccurate or biased outputs, automation bias, cybersecurity, and harms to health systems such as concentration of power among a few companies.

Audience	Examples of WHO recommendations
Governments	Invest in public infrastructure; use laws and regulation to require ethical and human-rights standards; assign a regulator to assess LMMs for health use; require post-release audits
Developers	Involve potential users, health professionals and patients from early in design; design for well-defined tasks with necessary accuracy; be able to predict and understand possible secondary outcomes
Health providers and users	Use LMMs only for appropriate tasks; keep clinical judgement; understand limitations and protect patient data

Regulation and research oversight

WHO's Regulatory considerations on artificial intelligence for health (2023) highlights documentation and transparency, risk management across the life cycle, intended use and validation, data quality, privacy and data protection, and engagement and collaboration.
In September 2026 WHO published a report on ethics review and oversight of AI-related health research, with recommendations for researchers, ethics committees, regulators and funders.
National instruments put similar principles into practice, for example the ICMR's 2023 AI ethics guidelines in India and the EU AI Act's requirements for high-risk systems.

Applying the principles to a decision

Principles help most when used on a specific question. Consider a hospital deciding whether to deploy an LLM tool that drafts replies to patients' portal messages for clinicians to review.

Principle	Questions for this deployment
Autonomy	Do patients know replies may be AI-drafted? Can they ask for a fully human reply?
Safety	How often are drafts wrong or unsafe, and do clinicians catch it? What happens with urgent messages?
Transparency	Is the tool's role documented and explained to staff and patients?
Accountability	Who signs the reply, and who investigates if something goes wrong?
Equity	Does it work as well for messages in other languages or from patients with low health literacy?
Responsiveness and sustainability	How will performance be monitored, and what does it mean for staff workload and skills?

Ethics and law reinforce each other

Many ethical principles are now legal duties in some places, such as transparency to users interacting with AI, data protection, and requirements for high-risk systems. Ethics guidance remains important where law is silent or still developing.

Exercise

Apply WHO's principles to a real-world style AI deployment decision.

Data: A fictional deployment proposal written for the notebook (Created for this course)

Read the fictional proposal for an AI triage chatbot in a diabetes clinic.
For each of the six principles, write two questions the proposal does not yet answer.
Identify which questions are also legal requirements in your country.
Draft three conditions you would attach to approval, and how you would check them after six months.

The runnable notebook and worked solution will be published soon. You can work through the steps with the code on this page in the meantime.

Key takeaways

WHO's six principles: autonomy, safety and well-being, transparency, accountability, equity, responsiveness and sustainability.
The 2024 LMM guidance addresses governments, developers and health providers.
WHO also covers regulation (2023) and ethics review of AI research (2026).
Principles are most useful applied to specific deployment questions.

Terms used: Automation bias, Human-in-the-loop, Large language model (LLM)

FAQ

What are WHO's principles for AI in health?

Protect human autonomy; promote human well-being, safety and the public interest; ensure transparency, explainability and intelligibility; foster responsibility and accountability; ensure inclusiveness and equity; and promote AI that is responsive and sustainable.

What does WHO say about ChatGPT-like tools in health?

Its 2024 guidance on large multimodal models sets out benefits and risks and gives recommendations for governments, developers and health providers, including regulatory assessment, inclusive design and appropriate use with human judgement.

Is WHO guidance legally binding?

No. It is guidance for countries and organisations. Many countries reflect its principles in their own laws and regulations.

How should a hospital use WHO's principles?

By turning each principle into specific questions about a proposed tool, such as who is accountable, how safety is measured and whether it works equally well for all patients, and making approval conditional on the answers.

Does WHO address AI in research?

Yes. In September 2026 WHO published recommendations on ethics review and oversight of AI-related health research.

Sources

Educational content only. This material teaches methods; it is not clinical advice. Code, models and examples here are not validated medical devices and must not be used to make decisions about real patients without proper clinical validation, governance and any required regulatory approval. Examples use synthetic or openly licensed de-identified data only. Never paste identifiable patient information into public AI tools. Mentions of models, tools or vendors are for identification, not endorsement.