Strategy meets reality: Healthcare AI in 2026
Ireland just published its first national AI strategy for healthcare. Three months earlier, every frontier AI lab launched Healthcare products in the same week. The timing tells us something important about where Healthcare AI stands right now: the supply side and the demand side are moving simultaneously but not necessarily in sync, writes John Ward, EY Ireland Partner, Head of Health Technology.
In March 2026, Minister for Health Jennifer Carroll MacNeill TD launched AI for Care covering 2026 to 2030, jointly commissioned by the Department of Health and the HSE. It builds on Digital for Care, the Department of Health’s digital health framework for Ireland covering 2024 to 2030 and the HSE’s corresponding Digital Health Strategic Implementation Roadmap.
Both are organised around six shared principles: patient as an empowered partner; workforce and workplace; digitally enabled and connected care; data-driven services; digital health ecosystem and innovation; and secure foundations and digital enablers. These principles established the digital foundation.
AI for Care now layers an AI-specific strategy on top, with four pillars: clinical care; operations; research and innovation; and public health. It also includes a three-horizon roadmap with 25 named priority opportunities, and a supporting AI implementation framework for EU AI Act compliance and HIQA oversight.
The horizon one priorities are pragmatic: AI-assisted imaging; clinical decision support; clinical documentation development and summarisation; demand and capacity management; contact centre and support function automation; research governance; data analysis and interpretation; and population health risk prediction. These are areas where AI is showing value around the world. The HSE has already estimated that over a million hours have been saved through automation and AI across finance and administration in the past five years.
It is a strategy grounded in what works, and that is a strength, but it also raises a question: which AI technologies will power this roadmap? Because the supply side is moving dramatically. In a single week in January 2026, all three frontier AI labs made major healthcare announcements, and while the headlines looked similar, the strategies diverge fundamentally.
OpenAI went direct with two propositions: one for consumers and another for healthcare providers. ChatGPT Health is direct to consumer. Connecting over 300 million people to medical records, Apple Health, and fitness apps. ChatGPT Health is for healthcare providers and gives clinicians a health insurance portability and accountability act (HIPAA)-compliant enterprise workspace with GPT-5 models evaluated using OpenAI’s HealthBench; 5,000 clinical scenarios scored against over 48,000 physician-written criteria. The model is the product and distribution is OpenAI-first.
Anthropic built connective tissue. Claude for Healthcare is an ecosystem play. It ships with pre-built connectors to CMS, ICD-10, NPI Registry, and PubMed’s over 35 million articles, plus agent skills for FHIR development, prior authorisation review, and care coordination. It is multi-cloud, available through AWS, Google, and Microsoft. The platform and how it integrates in your environment is the product.
Google open-sourced the foundation. Google’s MedGemma 1.5 is a free, multimodal model for medical imaging, clinical reasoning, and EHR interpretation. MedASR is a speech-to-text model based on the Conformer architecture that is pre-trained for medical dictation and transcription.
Both sit within Google’s Health AI Developer Foundations (HAI-DEF), a broader suite of open-weight models, tools, and resources designed to accelerate AI development for healthcare applications, requiring less data and computing than building from scratch. There is no packaged healthcare product; just open weights, a permissive licence, and a community ethos. The developer community is the distribution.
Where supply meets demand
AI for Care’s horizon one maps differently onto each frontier provider’s strengths.
The ‘imaging and clinical decision support’ priorities align closely with Google’s MedGemma; purpose-built for medical image interpretation across radiology, cardiology, dermatology, and endoscopy but it is not a turnkey, and it requires integration.
The ‘clinical documentation’ priority – ambient scribing, discharge summaries, record summarisation – is a domain where many choices exist whether that is Anthropic’s, OpenAI, Google’s MedASR. In addition to the frontier models, major digital health providers are embedding AI in their core such as Oracle’s Clinical AI Agent (already live in the UK NHS across Barts, Imperial, and Milton Keynes); so, depending on the existing technologies this may be the most immediately deployable.
The ‘research and innovation’ pillar is where Google’s open-source approach gives Irish academic partners the most flexibility to build on, and where Anthropic’s PubMed and ClinicalTrials.gov connectors offer researchers immediate access to a broad evidence base.
The ‘public health’ pillar, risk prediction, evidence synthesis, population screening, requires integration with the digital backbone that Digital for Care is building: the National Shared Care Record, the HSE Health App, and the forthcoming National EHR now in procurement. This is less about which frontier model is selected and more about whether the data foundation can be connected to the models in a secure, safe, and scalable way.
The real challenge
AI for Care gets the fundamentals right. It emphasises human-in-the-loop oversight, EU AI Act compliance, HIQA guidance, and the digital infrastructure prerequisites. The six guiding principles, person-centred; transparent and trustworthy; human-in-the-loop; lived experience; governance and safety; and proven benefit, are the right framework for a health system navigating this landscape. The strategy is grounded in the IPPOSI Citizens’ Jury recommendations and aligned to the WHO’s Ethics and governance of AI for health.
But frontier models are not waiting for implementation frameworks. They are shipping product now, and the gap between national strategy and vendor reality is where health systems either build genuine capability or accumulate disconnected tools that are difficult to govern; a scenario that is best avoided.
Ireland has the advantage of moving deliberately, with the EU AI Act as a regulatory foundation and Sláintecare’s integrated care vision as a guiding framework. The question is whether the governance structures – the AI and Automation Centre of Excellence, the intake mechanisms, the regional deployment pathways – can keep pace with a technology landscape that shifts quarterly.
The right response is not to chase every announcement. It is to build the strategic, technical, and cultural readiness to adopt the right models, for the right use case, at the right time. Writing a strategy is hard but making it a reality is harder.
That is the work ahead. EY is confident that there is the right alignment behind this strategy – political, systemic, and financial – and how it complements Digital for Care. That combination is rare in any health system. Ireland has a genuine opportunity to lead on responsible AI adoption in health, and we believe the foundations are there to make it happen.
John Ward, Partner
E: john.ward1@ie.ey.com
W: www.ey.com/en_ie
