Back To Blog

AI Ethics in Clinical Research: Common Principles Across U.S. and International Governance Frameworks

content contributor Tasha Mohseni
Tasha Mohseni, BS June 29, 2026
Table of Contents

Introduction

Artificial intelligence (AI) is increasingly being integrated into clinical research. Applications range from participant recruitment and protocol optimization to safety monitoring, data analysis, and predictive modeling. Investigators, sponsors, institutions, and oversight bodies face growing questions about how these technologies should be governed to support ethical research practices and participant protections. Numerous organizations have developed frameworks to guide the responsible development and use of AI. These frameworks were developed by different organizations, for different purposes, and with different audiences in mind. Some focus broadly on AI governance across sectors, while others are specific to health and medicine.

Although important differences exist among them, several common ethical themes emerge across these frameworks. This article will examine the following frameworks:

  • The World Health Organization’s (WHO) Artificial intelligence and evidence-informed policy: emerging challenges and opportunities: discussion paper
  • The U.S. National Institute of Standards and Technology (NIST) Artificial Intelligence Risk Management Framework (AI RMF)
  • The Organisation for Economic Co-operation and Development (OECD) Recommendation of the Council on Artificial Intelligence
  • The United Nations Educational, Scientific and Cultural Organization (UNESCO) Recommendation on the Ethics of Artificial Intelligence
  • National Academy of Medicine’s (NAM) Artificial Intelligence Code of Conduct for Health and Medicine: Essential Guidance for Aligned Action

Understanding the common ethical themes may help clinical research professionals navigate the evolving AI landscape while remaining grounded in longstanding research ethics principles. Note: Although this article focuses on widely recognized ethical AI frameworks, it is important to note that clinical research involving AI may also fall within the scope of regulatory authorities such as the U.S. Food and Drug Administration (FDA), particularly when AI systems meet the definition of a medical device or are used to support regulated decision-making. These regulatory frameworks are distinct from the ethical guidance discussed here but are an important part of the broader governance landscape.

Diverse Frameworks, Shared Ethical Questions

The WHO’s guidance focuses on the ethical and governance challenges associated with AI in healthcare and public health. UNESCO’s recommendation adopts a human-rights–based perspective. The OECD AI principles emphasize trustworthy AI and have influenced policymaking in many countries. NIST provides a practical framework for identifying and managing AI-related risks. The NAM’s AI Code of Conduct focuses specifically on responsible AI use in health, healthcare, and biomedical science.

Despite these preliminary differences, these frameworks address similar questions:

  • How should human oversight be maintained?
  • How can transparency and accountability be promoted?
  • How should organizations address bias and fairness?
  • What safeguards are needed to protect privacy?
  • How should risks be evaluated and managed?

Rather than representing a unified global approach, these frameworks offer complementary perspectives on common ethical challenges associated with AI.

Human Oversight

Human oversight is one of the most consistent themes across AI governance frameworks. The WHO guidance emphasizes that AI should support rather than replace human decision-making in health-related contexts. UNESCO and OECD similarly stress the importance of human agency and responsibility when AI systems influence outcomes that affect individuals and communities. NIST highlights governance structures that support accountability and ongoing oversight, while the NAM’s Code of Conduct emphasizes organizational responsibility throughout the AI lifecycle. In clinical research, AI systems may assist investigators in identifying participants, analyzing data, or generating insights. However, responsibility for participant welfare, scientific integrity, and regulatory compliance remains with human decision-makers. This emphasis on human oversight aligns closely with longstanding expectations regarding investigators’ responsibility to ensure adequate protections for human participants.

Transparency and Explainability

Transparency appears prominently across all five frameworks. WHO identifies transparency and intelligibility as essential characteristics of trustworthy AI. UNESCO and OECD similarly emphasize transparency as a mechanism for promoting accountability and trust. NIST encourages documentation and communication regarding AI system capabilities, limitations, and risks. The NAM’s framework highlights transparency as part of broader efforts to support quality, safety, and accountability. Transparency supports informed decision-making by investigators, sponsors, ethics committees, regulators, and research participants. Information regarding model development, validation, performance, and limitations may help stakeholders assess whether an AI system is appropriate for its intended use. While frameworks differ in how they define explainability, they generally recognize that stakeholders need sufficient information to evaluate AI-enabled decisions and processes.

Fairness, Equity, and Inclusion

Concerns regarding fairness and equity have become increasingly important in AI governance discussions. WHO guidance emphasizes inclusiveness and equity in AI for health. UNESCO highlights non-discrimination and fairness as core ethical principles. OECD identifies equitable outcomes as a component of trustworthy AI, while NIST encourages organizations to identify and manage harmful biases. The NAM’s Code of Conduct similarly promotes equitable access to AI benefits and systematic evaluation of bias. Clinical researchers may need to consider whether AI systems perform consistently across diverse populations and whether their use could inadvertently contribute to unequal treatment or participation opportunities. Although these frameworks generally agree on the importance of fairness, they differ in how fairness is defined, measured, or operationalized.

Privacy and Data Governance

Privacy and responsible data governance are foundational concerns across AI frameworks. WHO guidance emphasizes responsible stewardship of health data. UNESCO and OECD similarly highlight privacy protection and data governance. NIST addresses data quality, security, and governance as important components of AI risk management. NAM identifies data stewardship as a key element of trustworthy AI implementation. These principles are especially relevant in clinical research, where sensitive personal information is frequently collected, analyzed, and shared.

Risk Management and Continuous Monitoring

Another recurring theme is the need for ongoing evaluation and management of AI-related risks. NIST’s AI RMF is built around risk identification, assessment, and management. WHO guidance similarly emphasizes evaluating benefits and risks throughout the lifecycle of AI systems used in health contexts. OECD and UNESCO encourage organizations to anticipate potential harms and implement safeguards proportional to risk. The NAM’s framework highlights continuous monitoring, performance assessment, and organizational learning. A risk-based approach recognizes that not all AI applications warrant the same level of oversight. Continuous monitoring may help organizations identify unintended consequences and respond appropriately as AI systems evolve over time.

Important Differences Remain

Although common themes appear across these frameworks, important differences remain:

  • The WHO and NAM focus specifically on health and medicine, whereas UNESCO, OECD, and NIST address AI more broadly.
  • NIST emphasizes operational risk management, while UNESCO places greater emphasis on human rights and societal impacts.
  • OECD serves primarily as a policy framework, while NAM focuses on practical guidance for health-sector stakeholders.

The frameworks also differ in their intended audiences, implementation strategies, and governance objectives. For this reason, organizations should avoid viewing these frameworks as interchangeable.

Closing Thoughts

AI governance remains a rapidly evolving field, with multiple frameworks developed by regulatory agencies, international organizations, professional societies, and standards bodies. Although these frameworks differ in scope and purpose, several recurring ethical principles are present. For the clinical research community, these shared themes provide useful reference points for evaluating AI-enabled tools and processes while maintaining a focus on participant protection, scientific integrity, and public trust. Understanding both the similarities and differences among AI governance frameworks may help organizations navigate the expanding role of AI in clinical research while remaining attentive to the ethical responsibilities that have long guided human subjects research.

References

Related Content
Join Our Newsletter
This field is for validation purposes and should be left unchanged.
Related Articles
Recent Job Listings