We are pleased to announce a new open-access publication in Research Ethics, exploring an increasingly urgent question: as large language models move into clinical trial recruitment, can they communicate with prospective participants without quietly steering their decisions?
Informed consent rests on voluntary, well-informed choice that is free from undue influence. LLMs could genuinely help here — simplifying complex documents, translating materials, and answering questions on demand. But their generative, adaptive nature also makes their tone and framing hard to predict and control.
To make these dynamics visible, we ran an explorative case study based on a first-in-human phase I oncology trial. We prompted ChatGPT-5 to produce three consent texts under different instructions — standard, strictly neutral, and subtly nudged toward enrolment — and analysed them against established ethical categories: voluntariness, risks and discomforts, potential benefits, and scientific and social value.
The results show how much can change without changing a single fact. The nudged version emphasised intensive monitoring, reversibility, and the value of contributing to science, while omitting serious risks such as bleeding and QT prolongation that the standard and neutral versions disclosed. Small linguistic shifts were enough to move communication from autonomy-supportive information toward subtle persuasion — the kind of influence that, deployed at scale, could transform informed consent from a protective mechanism into a recruitment tool.
The paper concludes with five concrete safeguards: setting neutrality as the default communicative mode, formalising prompt governance and prohibiting recruitment-oriented framing, subjecting model tuning to independent scrutiny, ensuring transparency and auditability (with ethics committees documenting prompts, outputs, and safeguards), and explicitly mapping the design space of permissible interventions.
The work is a collaboration between the Institute of Data Science in Biomedicine and the Institute of Information Systems at TU Braunschweig and the Institute for Ethics, History and Philosophy of Medicine at Hannover Medical School (MHH).
Reference Rudra P, Balke W-T, Kacprowski T, Ursin F, Salloch S (2026). Algorithmic nudging for clinical trial participation: Autonomy and consent in the era of large language models. Research Ethics. https://doi.org/10.1177/17470161261451591 (open access, CC BY 4.0)