AI is making significant inroads into healthcare, from diagnostic imaging to patient prioritization. But its integration raises a question that goes beyond technology: How can we ensure that machines help improve medicine without replacing human judgment or jeopardizing patient safety?

A joint report by the European Academies Science Advisory Council and the Federation of European Academies of Medicine (FEAM), prepared by 22 European scientists, focuses precisely on this issue. The authors agree that AI in healthcare can bring benefits, but they caution that systems considered high-risk must remain under strict human supervision.

“AI should be a tool that empowers healthcare professionals, never one that replaces or displaces them,” said FEAM Vice President Luis Martí-Bonmatí, MD, PhD, who led the working group behind the report. He said the challenge has been turning that principle into recommendations that can apply across the 29 national healthcare systems in Europe, which differ widely in their levels of digitization, funding, and regulation.

Innovating Without Losing Control

One of the main debates has been how to strike a balance between regulation and innovation. Excessively rigid regulations can hinder the adoption of tools that provide clinical value, while insufficient regulation can leave risks related to safety, privacy, or automated decision-making unaddressed.

Among the report’s proposals are so-called regulatory sandboxes: controlled, time-limited environments where new AI tools can be tested under regulatory oversight, with safeguards for patients and, at the same time, allowing authorities to learn how to apply regulations to rapidly evolving technologies.

Security, moreover, must be built in from the design stage. David Ríos Insua, a working group member of the Royal Academy of Exact, Physical and Natural Sciences of Spain, and a contributor to the report, warned that AI systems may face adverse situations, including cyberattacks, and that their robustness must be assessed before they are deployed in clinical settings.

This concern is not merely theoretical. The growing integration of health data also increases exposure to cybersecurity incidents, while tampering with systems used for diagnosis could have clinical consequences. Added to this is the risk for bias: an algorithm trained on underrepresented populations may yield less accurate results for certain groups based on variables such as sex, age, or geographic context.

The report also warned of another, less visible risk: the gradual loss of professional skills, or deskilling. When an AI system routinely takes over certain tasks, healthcare professionals may lose the practice needed to perform them on their own. The authors believe this phenomenon must be taken into account because, in the event of a technology failure or outage, it could affect patient safety.

AI Also Needs Doctors

Training is another fundamental component of this transformation, according to the experts. The goal is not to turn doctors into AI specialists, but rather to ensure they understand what these tools can do, what their limitations are, and what risks are involved in using them in a healthcare setting.

The report also called for policies that ensure equitable access to these technologies, greater transparency regarding their operation and performance — including their environmental impact — and active public participation. The premise is that the value of AI depends not only on its technical capabilities but also on how it is developed, validated, implemented, and governed.

For professionals, the benefits can be tangible. In radiology, for example, AI can help detect lesions, reduce errors, prioritize studies based on clinical urgency, and streamline workflows. However, according to the report’s authors, none of these advantages eliminates the need for human judgment and responsibility.

From Control to Humanization

This concern for keeping the person at the center ties into a debate that transcends the strictly scientific realm. For years, the humanization of healthcare has advocated for care that does not reduce the patient to their illness but also takes into account their social, cultural, and emotional dimensions.

In 2026, this reflection also reached the Vatican with Magnifica Humanitas, Pope Leo XIV’s first encyclical, dedicated to the protection of the human person in the age of AI. The document argued that technology is not in itself an enemy of the person, but neither is it neutral: Its effects depend on who designs, funds, regulates, and uses it.

The encyclical calls for AI to be at the service of humanity and the common good and proposes criteria such as human dignity, social justice, solidarity, and the protection of the most vulnerable. Leo XIV also called for “disarming” AI — not in the sense of abandoning it, but of freeing it from the logic of domination, exclusion, and war.

This approach takes on a particular dimension when applied to healthcare. Magnifica Humanitas is not a document focused on healthcare, but its defense of the primacy of the person raises an issue related to the transformation of medicine: a technology capable of diagnosing, recommending, or assisting cannot reduce the patient to a set of data or the healthcare professional to a mere executor of algorithmic decisions.

Between the white coat and the Vatican, a shared concern thus emerges: that technological progress does not end up displacing what it seeks to protect. AI can transform the way we diagnose, treat, and organize healthcare, but its value will also depend on the continued presence of a person capable of interpreting its results, assuming responsibility, and caring for another person.

This story was translated from El Médico Interactivo on Univadis, part of the Medscape Professional Network.