Trustworthy AI in healthcare: A symphony of many voices

Trustworthy AI refers to AI technologies which are reliable, ethical, and accepted by society. In healthcare, where AI decisions could mean the difference between health and disease, or between life and death, the need for trustworthy AI is undeniable. Although it is recognised that trustworthy AI tools must avoid harm, maximise benefits, prevent discrimination, and be transparent, defining precisely the principles and methods to achieve these goals remains challenging.In 2025, after three years in the making, a scientific framework called FUTURE-AI was published [1], which established foundations for trustworthy AI in healthcare. This work, led by ICREA Professor Karim Lekadir, brought together 117 experts from various disciplines, including data science, clinical research, social sciences, and ethics. Participants from 50 countries across every continent contributed, to ensure the framework is suitable for worldwide application, including in resource-limited settings.Collaboratively, we established six core principles essential for trustworthy AI in healthcare: Fairness, Universality, Traceability, Usability, Robustness, and Explainability. Additionally, we outlined 30 best practices and a set of methods for aligning the AI tools with ethical standards, mitigating AI risks, collecting relevant training and testing data, generating evidence of trustworthiness, and ensuring post-deployment monitoring.A principal recommendation from our work is the continuous engagement of stakeholders from the very start. By hearing the voices of those that will be most affected by AI, especially patients, clinicians, healthcare providers, and citizens groups, we aim to co-create AI tools that are trustworthy by design.As an example, in the ERC-funded AIMIX project led by Professor Lekadir, we reached out to groups of pregnant women from rural Kenya (see photo) to understand their unique needs, hopes, and worries regarding AI-powered obstetric care. Multi-stakeholder engagement also revealed crucial pathways to adoption, such as through the participation of local religious leaders in the design and promotion of the AI tools to widen acceptance in these communities.