Experts Advocate for Patient-Centered AI Regulations in Healthcare to Prevent Discrimination

A recent commentary published in the Journal of the Royal Society of Medicine emphasizes the urgent need for a patient-centered approach to regulate artificial intelligence (AI) in healthcare. While AI and machine learning tools have the potential to improve clinical precision, current risk-based regulatory frameworks—like the European Union's AI Act of 2025—are falling short in safeguarding patients from over- and undertreatment as well as discrimination.

The authors highlight concerns over AI systems' inherent biases, lack of transparency, and potential inaccuracies which are not sufficiently addressed by existing regulations. The EU’s AI Act classifies medical AI as "high risk" and enforces strict controls on providers, but it overlooks critical issues such as individual patient preferences, the systemic and long-term impacts of AI deployment, and the disempowerment of patients in decision-making processes.

Lead author Professor Thomas Ploug from Aalborg University stresses that regulation must go beyond systemic safety and protect individual rights by incorporating patient participation. He advocates for establishing explicit patient rights in AI-driven care, including the right to explanations, to consent or withdraw consent, to seek second opinions, and to refuse diagnoses or screenings based on publicly available data.

The article warns that without proactive engagement from clinicians, regulators, and patient advocacy groups, these rights could be neglected amid rapid technological advancements. AI's influence in health care must be balanced with the preservation of patient autonomy and trust. The authors call for regulatory reforms that recognize these individual rights and ensure patients remain empowered in an AI-integrated healthcare system.

This discussion underscores the importance of evolving regulations to prioritize patient rights alongside technological innovation. For more details, see the full study: DOI: 10.1177/01410768251344707.