Clinical researchers testing artificial intelligence as a medical intervention now have a structured roadmap for navigating the ethical minefield that comes with algorithmic decision-making in patient care. A new framework developed by WCG, in collaboration with the MRCT Center and a multi-stakeholder group spanning bioethics, clinical research, regulatory science, and AI technology, provides institutional review boards (IRBs), sponsors, investigators, and research institutions with practical guidance on ethical oversight when AI is studied as a clinical intervention.

The tension is real: the same technologies that can accelerate medical care and improve decision-making also raise complex ethical questions that aren't always easy to anticipate or manage. For sponsors and contract research organizations (CROs), these challenges shape protocol design, participant protections, data governance strategies, and review readiness across the entire study lifecycle. Addressing them early can reduce avoidable delays, support more consistent IRB review, and strengthen confidence in study outcomes.

What Are the Six Core Ethical Considerations for AI in Clinical Research?

The framework identifies six key areas that IRBs and research teams must deliberate during the review process:

• Human Oversight: AI may support better, faster decisions, but it should not displace human judgment that affects diagnosis, treatment, disease prevention, or participant well-being. Researchers and clinicians must remain central to decision-making, with clear safeguards to prevent over-reliance on AI outputs or deference to algorithmic recommendations.
• Technical Robustness: Before AI is used with study participants, it should demonstrate that it is reliable enough for the role it is expected to play. Best practices include validation testing before deployment, performance monitoring during the study, and stronger oversight when algorithms continue to learn or adapt over time.
• Data Governance: Clinical research depends on data, often at a scale that can intensify privacy and confidentiality concerns. Strong data governance when AI is involved must include appropriate safeguards, documentation of training data and performance, and clear plans for retention and deletion.
• Transparency and Explainability: Reviewers, researchers, clinicians, and participants should be able to understand what the AI is meant to do, what its limitations are, and how it produces outputs that could influence care. This is especially important when black-box systems limit visibility into how conclusions are reached.
• Fairness and Representation: AI can only support fair research decisions if the data behind it is fit for purpose. IRBs should consider whether datasets are sufficiently representative of the population the AI will affect, and whether underrepresented groups may be excluded or disproportionately impacted.
• Informed Consent: The informed consent discussion must clearly outline the role of AI as the intervention in the research. Plain-language explanations of how the AI functions, how participant data will be used, and what privacy or confidentiality risks may exist are critical.

Why Does Data Representation Matter So Much in AI-Driven Clinical Studies?

One of the most practical concerns the framework raises is whether datasets used to train and validate AI systems actually reflect the real-world population the algorithm will serve. If the data doesn't match the population, bias isn't just a model problem; it can impact the AI output and ultimately the reliability of the intervention when deployed clinically across diverse individuals and populations. This raises questions about access, connectivity, and digital literacy, which could create inequities in who can participate in the research or benefit from the intervention.

The framework emphasizes that this is not a theoretical concern. For sponsors and CROs working across multiple data sources and study systems, governance needs to cover not only how data is protected but also how its provenance, traceability, and intended use are documented clearly enough to support ethical review and preserve participant trust.

How Should Sponsors and CROs Prepare for AI Ethics Review?

For organizations planning to test AI in clinical research, the framework suggests several practical steps to strengthen readiness and reduce delays during IRB review:

• Define Human Decision Points: Think beyond whether a model performs well. Define who reviews research outputs, when human intervention is required, how risks are assessed, and what study teams in training need to understand about the technology's limitations before it influences research or care.
• Document Technical Limitations: Be prepared to show how limitations are understood, how unexpected outputs or adverse events will be handled, and how ongoing performance will be monitored, especially if the system is adaptive. If an AI system changes in ways that affect participant risk, that may warrant additional review.
• Plan for Transparency: Transparency is not just a communications exercise; it is part of how risk is understood and how trust is built. Teams should understand the intended use of the AI, what information will be shared with participants, and how potential biases or errors will be identified and addressed.

The framework also highlights a growing reality in data governance: even when data is de-identified, it may still be linkable in ways that create re-identification risk. This has real implications for sponsors and CROs, particularly when working across multiple data sources and study systems.

Are These Concerns Really New, or Just Old Problems in New Clothing?

The ethical questions raised by AI in clinical research are often framed as entirely new challenges. In reality, many trace back to familiar research principles: protecting participants, minimizing risk, promoting fairness, supporting informed decision-making, and maintaining appropriate oversight. The framework helps translate those principles into practical review considerations for studies where AI acts as the intervention in the research.

For sponsors and CROs, the value of this lens is practical as much as ethical. Addressing these issues earlier can help teams surface avoidable risks, strengthen participant protections, improve the quality and clarity of protocol materials, and support a more thoughtful review process. As AI takes on a larger role in clinical care and is studied in clinical research, utilizing the ethical concepts outlined in the framework will help organizations ensure that the research meets appropriate criteria for participant protection and IRB review.

The bottom line: AI in clinical research is not inherently unethical, but it requires deliberate planning, clear documentation, and a commitment to keeping human judgment at the center of patient care decisions.