AI in Health: how smart technology is breaking new frontiers in medical care (deep-dive)

Artificial intelligence (AI) is no longer a future promise in European healthcare - it is already in radiology reading rooms, ophthalmology clinics, primary-care workflows and hospital back-offices. From CE-marked algorithms that flag suspicious findings on a mammogram in seconds, to AI systems that predict protein structures at unprecedented scale, AI-enabled tools are quietly breaking new frontiers in medical technology. This Digital Brief takes a grounded look at how AI is being deployed in healthcare for good, focusing on use cases where the benefits to patients, clinicians and health systems are tangible and measurable. It examines five domains - medical imaging, drug discovery, primary-care triage, public-health surveillance and clinician productivity - through the lens of the European policy framework that now governs them: the AI Act, the European Health Data Space (EHDS) and the Digital Decade 2030 targets. The brief also asks the harder questions: which clinical and digital skills will Europe’s health workforce need to use these tools safely; how do we keep humans meaningfully in the loop; and what does “AI for good” look like when the stakes are a patient’s life, dignity and data? It then closes with practical signposts to EU initiatives where readers can build the skills, partnerships and trust that will decide whether the next decade of AI in health is one of inclusion or exclusion.

Keywords

#Artificial intelligence in healthcare; #medical technology; #clinical AI deployment; #digital health skills; #trustworthy AI (EU AI Act).

Why this brief, and why now? 

Europe’s health systems are entering a decisive decade. Populations are ageing, clinical workforces are stretched, and citizens expect digital services in healthcare to match those they encounter in banking or travel. Against this backdrop, artificial intelligence (AI) has moved from research laboratories into routine clinical workflows rapidly, and across several clinical and administrative workflows at once. The Organisation for Economic Cooperation and Development (OECD) reports that AI is already supporting diagnosis, predicting patient deterioration, optimising hospital operations and accelerating biomedical research, with adoption visibly accelerating since 2023 (OECD, 2024). 

“Everyone should have online access to key public services in the EU... including and in particular digital health and care services, notably access to electronic health records.” 
- European Declaration on Digital Rights and Principles for the Digital Decade, Chapter II, point 7 (2023) 

Adoption, however, is not the same as benefit. The question that matters for European citizens - and for the workforce being asked to use these tools - is whether AI is delivering tangible, measurable good. This brief focuses on five domains where the evidence is strongest and where European policy is already shaping deployment: medical imaging, drug discovery, primary-care triage and access, public-health surveillance, and clinician productivity. Throughout, it situates these advances within the European policy stack - the EU AI Act (Regulation (EU) 2024/1689), the European Health Data Space (EHDS) and the Digital Decade 2030 Policy Programme - so readers see not only what is possible, but what is permissible, and which skills the European health workforce will need to make it work.

How is AI already changing medical technology? 

Imaging: can an algorithm read a scan as well as a radiologist? 

Medical imaging is among the most mature clinical applications of AI. The United States Food and Drug Administration’s public list of AI-enabled medical devices, while not a European dataset, provides a useful scale signal for how quickly this category is growing; in Europe, comparable tools enter clinical use through MDR/IVDR conformity-assessment pathways and CE marking (FDA, 2025). Prospective evidence is accumulating. The Swedish MASAI randomised controlled trial, published in The Lancet Oncology, compared AI-supported double reading of mammograms with standard double reading by two radiologists across more than 80,000 women, and found AI support detected around 20% more cancers without increasing recall rates, while almost halving radiologist screen-reading workload (Lång et al., 2023). In ophthalmology, European policy and technology-assessment literature frequently identifies diabetic-retinopathy screening as a clear example of how AI can extend specialist-level diagnostic capacity into primary care and underserved regions (Lekadir et al., 2022). 

What is breaking frontiers here is not that AI replaces clinicians - it does not - but that it changes the economics of expertise. A retinal photograph taken by a nurse in a rural pharmacy could be triaged in seconds with sensitivity comparable to a specialist. A subtle pulmonary nodule a tired radiologist might overlook at the end of a long shift can be flagged for a second look. Scaled across screening programmes, such gains could improve earlier detection and workforce capacity, provided deployment is continuously monitored for safety, equity and false positives. The workforce consequence is real: radiologists are moving from primary-reader to AI-supervisor roles, and primary-care nurses are taking on new triage responsibilities - changes that reshape which digital and clinical skills the European health workforce must be trained in. 

Drug discovery: can AI shorten the road from molecule to medicine? 

Bringing a new medicine to patients has historically taken many years and required substantial investment. AI is beginning to compress parts of that pipeline. The most cited example is AlphaFold, developed by Google DeepMind and the European Molecular Biology Laboratory’s European Bioinformatics Institute (EMBL-EBI), which has predicted the three-dimensional structure of more than 200 million proteins and made them openly available (EMBL-EBI, 2024). The 2024 Nobel Prize in Chemistry recognised this contribution, underlining how foundational AI has become to biomedical science (Nobel Prize, 2024). Structure prediction does not by itself produce safe or effective medicines, but it changes the starting conditions for biomedical discovery. Beyond structure prediction, generative AI is being used to propose novel drug candidates, optimise clinical-trial design and re-purpose existing compounds for rare diseases - areas where commercial incentives have historically been weak. 

For European citizens, the practical benefit is twofold. First, accelerated discovery raises the probability that effective treatments emerge for conditions that today have none, including many paediatric and rare diseases. Second, Europe’s strong public-research base - embodied in Horizon Europe (the EU’s key funding programme for research and innovation) and the EHDS - gives the Union a credible path to capture some of this value within its own innovation ecosystem (European Commission, 2026b). That depends on the availability of high-quality, lawfully shared, well-curated health datasets, which is precisely what the EHDS is designed to enable (European Commission, 2025a). The skills implication is equally important: demand is rising for clinical bioinformaticians, computational chemists and AI-fluent translational researchers across European pharma and academia. 

Primary care: can AI extend access to those the system reaches last? 

Some of the most consequential AI deployments are not in tertiary hospitals but at the front door of the health system. Symptom-checker chatbots, AI-supported triage in emergency departments and ambient-listening tools that transcribe and structure a consultation are moving from pilots into operational use in parts of Europe, although evidence quality and deployment maturity vary. The OECD’s 2024 review notes that such tools, when well governed, can shorten waiting times, reduce administrative burden on general practitioners and improve continuity of care for chronic conditions (OECD, 2024). The World Health Organisation’s guidance on AI ethics in health emphasises the same potential, while warning that poorly validated chatbots can entrench inequities if trained primarily on data from high-income, English-speaking populations (WHO, 2024). 

Europe’s comparative advantage here is regulatory clarity. Under the EU AI Act, many clinical AI systems will be treated as high-risk either because they are products, or safety components of products, covered by Union harmonisation legislation listed in Annex I - such as the MDR and IVDR - or because they fall within specific standalone high-risk areas listed in Annex III (European Parliament and Council, 2024). For citizens, this means that an AI tool used for clinical diagnosis, triage or decision support may be subject to obligations that go well beyond those applying to a consumer wellness app. The job effect is significant: GPs and triage nurses will increasingly need AI-literacy and override skills as part of their core competence, while new roles - digital-health coordinators within practices, oversight officers in larger groups - are emerging to keep human oversight real rather than ceremonial. 

Public health: can AI help us see outbreaks before they spread? 

The COVID-19 pandemic accelerated investment in AI-enabled public-health surveillance, and the lessons are now being institutionalised. European public-health intelligence now combines indicator-based surveillance, event-based surveillance and open-source monitoring. Tools like EpiPulse of the European Centre for Disease Control and Prevention (ECDC) support the collection, analysis and sharing of infectious-disease signals across Member States, while WHO-led EIOS workflows use open-source information and technology-enabled monitoring to help detect emerging public-health threats earlier (ECDC, 2021; WHO EIOS, 2025). When combined with machine-learning models, wastewater signals, genomic surveillance and hospital-admissions data can support earlier warning and resource planning. These applications illustrate a quieter face of AI for good: not the dramatic single diagnosis, but the population-scale early warning that buys clinicians and governments time. 

Similar techniques are being applied to non-communicable disease. Models that integrate electronic health records, environmental data and social determinants can identify communities at elevated risk of cardiovascular disease, diabetes complications or mental-health crises, enabling targeted preventive outreach. Done well, this is precision public health; done badly, it risks reinforcing biases embedded in routine data. The governance framework matters as much as the model - and so does the workforce: data-engineering, geospatial-analytics and epidemiological-modelling skills are shifting from optional to core in national public-health agencies across the Union. 

Clinicians and operators: can AI give time back to the people who heal? 

Burnout is now a structural risk for European health systems. Many of the most welcomed AI deployments are therefore not those that try to do clinical reasoning, but those that absorb administrative burden - ambient scribes that draft consultation notes, intelligent rostering tools, predictive bed-management systems and automated coding for reimbursement. These tools rarely make headlines, yet they may have the largest near-term impact on the lived experience of working in healthcare. Early deployments suggest that ambient documentation, intelligent rostering and predictive bed-management tools can reduce administrative burden, but their impact depends heavily on workflow redesign, procurement quality and clinician co-design. The job effect is twofold: administrative roles evolve rather than disappear, and hospital managers need procurement and AI-governance literacy to deploy these tools safely. Both the European Commission’s European Year of Skills follow-up actions and the Digital Decade targets recognise that upskilling the workforce to use these tools confidently is as important as procuring them (European Commission, 2024).

What does the European policy framework actually require? 

Three instruments deserve particular attention. First, the EU AI Act establishes a horizontal, risk-based legal framework for AI across the Union. Many clinical AI systems will be treated as high-risk either because they are products, or safety components of products, covered by Union product-safety legislation such as the MDR and IVDR (Article 6(1) and Annex I), or because they fall within specific standalone high-risk use areas listed in Annex III. This triggers requirements on risk management, data governance, technical documentation, transparency, human oversight, accuracy, robustness, cybersecurity and post-market monitoring. The implementation timeline is now phased: the AI Act is generally applicable from 2 August 2026; following the political agreement on the AI omnibus simplification proposal, rules for standalone high-risk systems under Annex III apply from 2 December 2027, and high-risk rules for systems embedded into regulated products have an extended transition period until 2 August 2028 (European Parliament and Council, 2024; European Commission, 2026a). 

Second, the European Health Data Space Regulation (Regulation (EU) 2025/327) entered into force on 26 March 2025 creating a Union-wide framework for the primary and secondary use of electronic health data, strengthening citizens’ rights, and opening a governed pathway for researchers and innovators to access pseudonymised data - including to train and validate AI systems. Because EHDS implementation is phased, its operational value for AI training, validation and secondary research use will depend on Member State implementation, Health Data Access Bodies, secure processing environments and interoperability rules becoming operational over the coming years, with key parts applying from March 2029 and further categories from March 2031 (European Commission, 2025a). Third, the Digital Decade 2030 Policy Programme sets concrete targets for 2030 - including that at least 80% of adults should have basic digital skills, alongside the Digital Decade eHealth target of providing all EU citizens with access to their electronic health records - framing the demand side for digital health skills (European Commission, 2025b; European Commission, 2025d). 

Around these three core instruments sits a wider European policy ecosystem that shapes how AI in health is funded, governed and made trustworthy. The EU4Health Programme 2021-2027 supports stronger, more resilient and more accessible health systems, including digital transformation measures that can complement AI-enabled healthcare innovation (European Commission, 2025c). The European Strategy for Data underpins the broader data infrastructure on which the EHDS rests, by promoting sectoral data spaces and trusted cross-border data exchange (European Commission, 2020). The European Declaration on Digital Rights and Principles gives citizens a politically endorsed reference point on online access to digital health and care services, human-centric AI and meaningful oversight (European Parliament, Council and Commission, 2023). And the EU’s Cybersecurity Strategy for the Digital Decade is highly relevant here too: with healthcare exposed to serious cyber incidents across Europe, any AI system handling patient data must be defensible by design as well as compliant on paper (European Commission, 2020b; ENISA, 2024). 

Together, these instruments mean that in Europe the question is no longer whether AI will be regulated in healthcare, but how organisations, clinicians and patients will operationalise that regulation. That, in turn, is fundamentally a skills question. 

Which digital skills will Europe’s health workforce need? 

If the policy frame is clear, the skills frame is still under construction. Drawing on the European Digital Competence Framework for Citizens (DigComp 3.0, the fifth edition of the framework, which systematically integrates AI across its dimensions), the European e-Competence Framework (e-CF) and emerging national curricula, four skills clusters can be identified as essential for the safe and ethical use of AI in health (Cosgrove and Cachia, 2025; CEN, 2019). These are summarised in Table 1 below, with an indication of which parts of the workforce most need each cluster. 

Table 1. Core AI and digital-health skills for Europe’s health workforce. Source: author’s own synthesis, based on DigComp 3.0 (Cosgrove and Cachia, 2025), OECD AI-in-health analysis (OECD, 2024), EU AI Act obligation. 

Two practical implications follow. First, AI skills cannot be confined to data scientists or chief information officers; they must be embedded across professional curricula. Second, the right starting point for most professionals is not a coding course but a structured AI-in-practice literacy programme grounded in real clinical scenarios. EU instruments such as the Digital Skills and Jobs Platform, the Pact for Skills health and care ecosystem and the European Year of Skills follow-up actions are designed to make this kind of provision easier to find, coordinate and scale (European Commission, 2024a; European Commission, 2024b; CEN, 2019).

The infographic can be viewed in full, in PDF format via this link. 

Where could this go wrong - and how do we keep AI “for good”? 

A brief that is honest about AI for good must also be honest about AI for harm. Three failure modes deserve naming. The first is performance drift: a model that performs well at launch can quietly degrade as clinical practice, patient demographics or upstream data pipelines change, with consequences that are clinically opaque unless monitoring is genuinely operational. The AI Act’s post-market monitoring obligations address this in principle; operationalising them is a skills and infrastructure challenge in practice. 

The second is inequity. Models trained on data from one part of Europe, or from one demographic group, can underperform on others. Diabetic retinopathy models trained primarily on lighter-skinned populations, or risk scores derived from health systems with different referral patterns, may systematically disadvantage some patients. Mitigation requires diverse training data, disaggregated evaluation, and the kind of cross-border, lawfully governed data sharing the EHDS is intended to enable (European Commission, 2025a). 

The third is erosion of the clinical relationship. Patients place trust in clinicians because clinicians take responsibility. AI can support, but cannot displace, that locus of responsibility. The EU AI Act enshrines human oversight as a legal requirement for high-risk systems; the harder cultural work is ensuring that oversight is real rather than ceremonial - that clinicians have the time, training and institutional backing to disagree with a model when they should. 

A fourth, cross-cutting risk is cybersecurity. An unsafe AI system in healthcare is not only one that performs poorly clinically, but one whose data flows, model updates and clinical interfaces can be attacked. With healthcare exposed to serious cyber incidents and ransomware risks across Europe, defensible-by-design AI deployment is no longer an information-security afterthought but a precondition for trustworthy AI in health (European Commission, 2020b; ENISA, 2024). Done well across all four risks, this is what “AI for good” looks like in healthcare: not the absence of humans, but the strengthening of human judgement, with the technical and organisational safeguards to back it up.

What can readers do next? 

• Build your own AI literacy through structured EU-aligned resources such as the Digital Skills and Jobs Platform and the courses and competence frameworks it indexes, including DigComp 3.0. 
• Engage with the European Health Data Space implementation in your country to shape how secondary use of health data - including for AI - will work in practice. 
• If you procure, deploy or use clinical AI, map your obligations under the EU AI Act against the current implementation timeline, including the extended transition period for high-risk systems embedded into regulated products. 
• Embed AI cases into clinical education using open guidance, policy briefs and evidence resources published by the WHO and the OECD. 
• Champion equity and security from the start - insist on disaggregated performance data, diverse training populations, meaningful patient involvement, and cyber-secure deployment in the design and evaluation of clinical AI.

Closing reflection

AI is breaking frontiers in medical technology not principally because the algorithms are clever, but because Europe is building the legal, ethical and skills infrastructure that allows clever algorithms to be used safely at scale. The frontier that matters most over the next five years is therefore not technical. It is the frontier of trust - between citizens and their data, between clinicians and their tools, and between Member States co-creating a Health Union worthy of the name. The Digital Skills and Jobs Platform exists precisely to help everyone, from a hospital director to a nursing student to a regional policy-maker, cross that frontier with confidence.

View this paper in full, together with its references via this link. The infographic is also available in PDF format - click here to access it. 

About the author

Roy Saurabh is the Founder and CEO of AffectLog, an EU-based deeptech company working on trustworthy AI governance and assurance for regulated sectors, including healthcare. His work focuses on privacy-preserving computation, regulatory reasoning and human-centred AI oversight for high-risk systems. He has contributed to European and international work on AI governance, digital skills and data spaces, and writes on the intersection of artificial intelligence, public-interest technology and digital regulation.

About the organisation

AffectLog is building privacy-preserving infrastructure for trustworthy AI in regulated sectors, including healthcare. Its work centres on enabling organisations to derive insight from sensitive data without exposing the underlying records, and on operationalising AI governance obligations (such as data protection, human oversight and auditability) in real-world deployments. AffectLog engages with European initiatives on health data, digital skills and trustworthy AI, and contributes to the wider effort to make advanced AI safe, accountable and accessible across the Union.