Ethical challenges and mitigation strategies in LLM use in health research: perspectives of research ethics committees in South Africa

The integration of large language models (LLMs) into health research offers substantial opportunities for hypothesis development, data synthesis, and manuscript preparation. However, these tools also raise significant ethical concerns, particularly regarding privacy, bias, plagiarism, and authorship. Research Ethics Committees (RECs) play a critical role in overseeing such research, yet little is known about how REC members and staff perceive these challenges in low- and middle-income country contexts. This qualitative study explored the perspectives of REC members and Research Ethics Office (REO) staff from institutions across South Africa using semi-structured in-depth interviews. Between January and July 2024, interviews were conducted with 14 respondents who had been selected using purposive and snowball sampling techniques. Thematic analysis identified two overarching themes. The first is perceived ethical challenges – privacy and data protection (including contravention of the Protection of Personal Information Act, POPIA), algorithmic bias (especially underrepresentation of African populations and human-prompt-driven bias), and plagiarism or authorship ambiguity. The second are strategies for responsible use – transparency and accountability, engagement of key stakeholders (RECs, institutions, researchers), and the need for institutional AI guidelines. Respondents emphasised that eliminating bias is likely impossible, but transparency in LLM use and formal acknowledgement can mitigate harms. Also, guidelines provide the floor; ethical legitimacy is earned beyond it and accountability to those who never consented and will still bear the consequences. The findings highlight the urgent need for context-sensitive governance frameworks that incorporate and priorities capacity building for RECs in AI ethics, and institutional agreements with LLM providers. This study provides empirical evidence to inform policy development and ethical oversight of LLMs in health research in the Global South.