How AI is quietly reshaping Spanish healthcare
Artificial Intelligence is no longer a future prospect in Spanish healthcare. It is already examining patients, prioritising cases, and shaping diagnoses, often without patients’ explicit knowledge. This investigation for Civio traces how AI systems are being introduced across public hospitals, focusing on ophthalmology as a revealing test case, and asks whether putative technological efficiency is coming at the cost of transparency, actual effectiveness, and human judgement.
The article opens with a first-person encounter at Madrid’s La Paz University Hospital, where one of the authors undergoes an eye exam conducted almost entirely by an AI-powered diagnostic booth called Eyelib. In under ten minutes, the system takes more than a hundred measurements using automated instructions. Only afterwards does the patient learn that the exam relied on AI. Eyelib, distributed under the brand DORIA, has cost the hospital over €1.2 million through public contracts and has been operating in several specialist centres since early 2024.
Hospital administrators argue that such systems are a necessary response to crisis conditions. Waiting lists in ophthalmology ballooned after the pandemic, and Eyelib was introduced to process large volumes of patients quickly. Yet official data from the Madrid Health Service undermines this promise. Since Eyelib’s introduction, both patient numbers and waiting times at La Paz have slightly worsened, raising doubts about whether automation is solving the problem it was meant to address.
Medical professionals themselves are divided. Some acknowledge the technical thoroughness of AI-driven exams but question their efficiency and cost. Critics note that each AI-generated report must still be reviewed and signed off by a specialist, pushing costs higher rather than lowering them. In Eyelib’s case, estimates place the cost per patient at around €80, before human oversight is factored in. Others point out that the system often sees fewer patients than a trained ophthalmologist working directly.
Beyond ophthalmology, the article maps the rapid expansion of AI across Spain’s health system. Algorithms are now used to analyse medical images in radiology and dermatology, assist in radiotherapy planning, manage surgical workflows, prioritise emergency patients, and even conduct follow-up calls with cancer patients via conversational chatbots. Some of these tools are publicly developed, while others are supplied by private companies in partnership with pharmaceutical firms, blurring the line between public care and outsourced technology.
Supporters argue that AI can genuinely improve outcomes. Faster imaging could reduce the need to sedate children during scans. Early detection tools may catch diseases before symptoms worsen. Health officials insist these systems are meant to complement, not replace, medical professionals.
The article highlights a core technical risk: algorithms can appear highly accurate while relying on flawed correlations. Researchers recount how a system trained to detect COVID-19 pneumonia from X-rays initially seemed nearly perfect, until they discovered it was “diagnosing” based on whether patients were hunched over, not on lung pathology. Such errors, if unnoticed, could have serious consequences in real clinical settings.
Other failures have already occurred. A medication chatbot launched by Spain’s medicines’ agency was suspended after giving nonsensical answers, recommending unsafe dosages, and confusing everyday phrases with medical advice. A separate AI system designed to help diagnose rare diseases failed entirely when tested with a real patient’s long-documented condition. These cases illustrate how poorly supervised tools can create new risks rather than reduce existing ones.
Finally, the article confronts the unresolved question of responsibility. If an algorithm makes a harmful mistake, who is accountable? European rules require human oversight of high-risk AI systems, but in practice, professionals are often asked to validate machine decisions after the fact. This creates pressure to defer to the algorithm rather than challenge it. While compensation is theoretically possible, consumer advocates argue that patients are unlikely even to know AI played a role in their care.
The piece paints a picture of a healthcare system embracing AI faster than it can regulate, understand, or properly scrutinise it. The promise of efficiency is real, but so are the dangers of opacity, overconfidence, and misplaced trust.
This article was originally published in Spanish as ‘Cuando tu médica es una IA’ by Ángela Bernardo and María Álvarez del Vayo in Civio on 9 October 2025.
It is available here.
Civio is an independent Spanish nonprofit investigative journalism organisation focused on transparency, public accountability, and data-driven reporting.
