Home » Insights » AI Health Research Ethics: What WHO’s New Report and Recent African Studies Mean for SA Psychology

AI Health Research Ethics: What WHO’s New Report and Recent African Studies Mean for SA Psychology

The World Health Organization (WHO) has released new guidance on AI health research ethics, and the timing matters for South African psychology. On 21 September 2026, WHO launched Artificial Intelligence-related Health Research: Ethics Review and Oversight, a report aimed at researchers, ethics committees, regulators and funders (World Health Organization, 2026). Meanwhile, three recent African studies show how these questions already play out on our continent. If you sit on a research ethics committee, supervise student research or use AI tools in practice, this affects you. For more on ethics in practice, browse our psychology insights.

What the WHO report on AI health research ethics covers

WHO describes the report as a set of practical recommendations for the responsible conduct of AI-enabled health research. The launch discussion centred on four themes: informed consent, bias in AI-related health research, the capacity of research ethics committees, and risks in low-income settings (World Health Organization, 2026).

Each theme lands differently in South Africa. Protocols that involve algorithms, chatbots or automated scoring raise questions that traditional review templates were never designed to answer. Consequently, committees need new questions, not just new forms.

Consent when the algorithm changes

Informed consent assumes participants understand what they are agreeing to. AI complicates that assumption. For instance, developers may retrain or update a model partway through a study, so the tool a participant agreed to is no longer quite the same tool. Ethics committees therefore need to ask how researchers will track changes and when re-consent becomes necessary.

Bias built into the tool

AI systems learn from their training data, and data drawn from other populations may not reflect South African patients. As a result, a tool can perform well where it was built yet mislead here. Mussie (2026), discussed below, notes that datasets which underrepresent African populations are a particular concern once these systems enter African clinics.

South African ethics committees: AI should assist, not decide

A South African study in BMC Medical Ethics offers the clearest local signal. Obasa and Kabanda (2026), from Wits and Stellenbosch University, interviewed 14 research ethics committee chairs, members and ethics office staff at health science institutions across the country.

Participants saw real benefits in large language models (LLMs), including administrative efficiency and support across the research lifecycle. Nevertheless, they agreed unanimously that LLMs should augment ethics review and never automate it. Specifically, they considered human oversight irreplaceable for contextual understanding, empathy and complex moral deliberation. The authors link these values to relational, Ubuntu-informed ethics.

The study also identified fragmented AI literacy in research settings. In other words, the people who guard research integrity do not yet share a common understanding of the technology. Keep in mind that the interviews took place in 2024, so views may have moved since.

Lessons from TRACE: fix the process, not only capacity

Capacity problems are not only about knowledge. They are also about process. The TRACE initiative (Trial Regulation and Clinical Ethics Optimization) works across Rwanda, Tanzania, Nigeria, Zimbabwe and Kenya with national ethics committees, institutional review boards and regulators (Neil et al., 2026).

Writing in Nature Medicine, the TRACE group reports a telling pattern. Across all five countries, the fastest and most measurable gains in trial readiness came from running ethics and regulatory review as one coordinated, parallel process. In contrast, adding technical capacity in isolation delivered less.

The authors also point out that Africa carries nearly a quarter of the global disease burden but hosts only about 2% of the world’s clinical trials (Neil et al., 2026). Slow or duplicated review therefore carries real costs for African patients and researchers.

Ethical reasoning, consent and professional judgement sit at the heart of good practice. Each of our HPCSA-accredited CPD packages includes 6 ethics CEUs alongside its clinical content.

Why AI health research ethics extends into the consulting room

The final study moves from research into clinical care. In Science and Engineering Ethics, Mussie (2026) of Addis Ababa University examines AI clinical decision-support tools in African health settings. The warning is subtle. These tools can keep formal human sign-off in place while quietly shifting authority, responsibility and the standards of good judgement away from the relationships where care happens.

Mussie calls this “relational displacement”. Time pressure and staff shortages can make deferring to an algorithm the rational choice, even when a practitioner has reasons to doubt it. Over time, the clinician’s role risks becoming procedural rather than substantive.

For psychologists, this should sound familiar. Our work depends on context, narrative and relationship, which are exactly the elements an algorithm struggles to encode. Accordingly, Mussie proposes a relational impact assessment that evaluates how AI tools affect trust, interpretive agency and family involvement, not only accuracy.

What this means for South African psychology professionals

Locally, the HPCSA published a revised Booklet 20 on the ethical use of AI in late 2025, and the South African Medical Association’s AI task team has since reflected critically on it (Mahomed et al., 2026). Taken together with the research above, a few practical priorities stand out:

  • If you serve on an ethics committee: ask how any AI tool in a protocol will be monitored, updated and explained to participants.
  • If you supervise research: discuss where students use LLMs and how they should disclose that use.
  • If you use AI tools in practice: treat outputs as input to your judgement, and document your reasoning when you depart from them.
  • For everyone: build your AI literacy, since the local evidence suggests understanding remains uneven even among ethics gatekeepers.

Finally, none of these studies argues against AI. Instead, they argue for keeping human judgement meaningful rather than merely formal.

Staying current on AI health research ethics is now part of practising responsibly. Explore our 2026 CPD packages to build your ethics CEUs, and if you graduated recently, check whether you qualify for our recent graduate discount.

References

Shopping Basket
Scroll to Top