In less than a decade, artificial intelligence (AI) has become key to developing new models of business and society. This ambitious trend is particularly evident in the healthcare sector, which is facing a continuously rising demand due to an ageing population, an increase in chronic and complex conditions, the high costs of innovative treatments, and frequent shortages in the healthcare workforce.
The potential impact of artificial intelligence on healthcare has been thoroughly analysed by the more than 200 pages document “Study on the deployment of AI in healthcare”, published by the Commission’s Directorate-General for Health and Food Safety (DG Santé) in August 2025.
The depicted scenario provides many opportunities for improvement, such as enhancing operational efficiency, reducing administrative burdens, and optimising diagnosis and treatment pathways. A specific case study is also presented for each topic in the Annexes. Nevertheless, despite the increasing availability of AI-based tools on the market, there are still many challenges to ensure the swift deployment of AI in healthcare. These include technological and data-related issues, legal and regulatory complexities, organisational and business challenges, and social and cultural barriers. We summarise the main ones, along with with the strategies suggested to overcome them.
Improved management of the increasing demand
Artificial intelligence could play a key role in optimising operational efficiency and improving the overall sustainability of healthcare systems. According to the Commission’s study, for example, AI could assist with the allocation of hospital staff resources (e.g. by helping with patient triage) and improve the prediction of patient flow and service demand by identifying those likely to require intensive care or longer hospital stays, for example.
Artificial intelligence may also contribute to reducing the administrative burden that healthcare provides (HPCs) face due to the implementation of electronic health records (EHRs). AI systems based in Large Language Models (LLM) and Natural Language Processing (NLP), for example, could help HCPs with non-clinical tasks such as documenting encounters, back-office functions, and patient scheduling. The report mentions as an example digital scribes, which combine speech recognition with NLP to automate clinical documentation and enhance data accuracy.
The potential of AI in diagnosis and treatment
Delayed diagnosis, disease progression and reduction of treatment effectiveness are three strictly correlated elements which may benefit from the potential offered by AI systems. In assisted diagnosis, for example in radiology and digital pathology, they could be used to improve the speed and accuracy, while also reducing the variability between HCPs responsible for interpreting diagnostic results.
AI systems that support clinical decision-making (CDSS) are also being used increasingly in many hospitals, thanks to their ability to rapidly analyse large amounts of data of each patient’s disease history, including medical history, genetic information and lifestyle factors. AI-driven robotic systems have become the norm in many surgical departments, where they are used to analyse pre-operative images more effectively for surgical planning, to improve the precision of instruments used during the surgery and to predict the possible complications.
The same considerations also apply to the highly challenging sector of cancer care to a greater extent. AI finds here a number of potential applications, including advanced screening techniques that allow for the early detection of tumors with greater sensitivity and specificity than traditional methods. Comparing images taken at different times may also help predict the appearance of metastasis, for example.
However, the real game changer could be represented by the use of AI-powered CDSS to assist the selection of the most appropriate therapy for each patient, thereby optimising outcomes from both perspective of both patients and healthcare systems.
The regulatory perspective
The regulatory perspective is one of the more challenging aspects of deploying AI in healthcare, due to the need to comply wit many complex and often overlapping vertical and horizontal pieces of legislations.
Cross-sectorial regulations obviously include the AI Act, the recent European regulation that governs the entire framework of AI development, implementation and use according to a risk-based approach.
Artificial intelligence systems also fall under the Product Liability Directive (PLD), which sets out how victims should be compensated if they are harmed by defective products, including AI systems. From this point of view, AI systems are characterised by a high degree of complexity, opacity, and autonomous capabilities. The PLD establishes compensation for harm even when the defect cannot be attributed to a specific fault.
Artificial intelligence also falls under the Medical Devices (MDR) and In Vitro Diagnostics (IVDR) sectorial regulations. More specifically, many AI systems used for healthcare applications (i.e. to assist with diagnosis or treatment) are considered “software as a medical devices”, and must therefore be developed and managed throughout their entire life cycle in accordance with the relevant MDR/IVDR rules, in order to ensure that they are safe for the patients/users and perform correctly. Post-market surveillance and reporting obligations also apply.
As AI is often incorporated into medical devices, their effectiveness and safety may be subject to scrutiny under the Health Technology Assessment Regulation (HTAR), for example through activation of joint clinical assessments of high-risk medical devices.
The European Health Data Space (EHDS) will progressively include an increasing amount of healthcare data from European patients. AI tools will be essential for analysing these data, providing different types of analyses and achieving the integration of healthcare systems at the European level. From this perspective, interoperability is a major challenge to be faced also in the deployment of AI systems.
The main issues in the deployment of AI in healthcare
The Commission’s report highlights that research on AI in healthcare increased after 2019, peaking in 2022 with 85 studies reported in the literature. The total budget of the monitored projects in the period 2015-2024 was approx. €3.53 billion. There has also been a constant increase in patenting activity and clinical trials involving AI systems. For example, the FDA lists a total of 950 AI/ML-enabled medical devices approved up to June 2024.
Despite these encouraging findings, the adoption of AI medical devices in the clinical practice is still limited, according to the report. Furthermore, the analysis indicates that most AI/ML-enabled devices are classified as medium/low-risk products; therefore, the human component remains predominant in higher–risk clinical operations and interventions.
The report identifies the lack of data standardisation and interoperability, the absence of new infrastructures and validation protocols and the lack of transparency and explainability as among the main challenges to be faced for the implementation of AI in healthcare. Ensuring the security and privacy of data is one of the regulatory and legal issues that need to be addressed, along with the lack of frameworks for liability and accountability.
From the operational and business perspective, specific funding and financing mechanisms would be required. The report also mentions the lack of strategic direction, added-value assessment and end-user involvement. Trust among patients and the general public should also be improved, as well as the level of digital literacy. The report also raises some concerns about job security, overreliance on artificial intelligence and its impact on the patient-doctor relationship.
