Most conversations about AI in healthcare still begin with the model: what it can do, how accurate it is, how reliable. Practitioners have moved on. The question I hear from them is this: how does an AI system enter an operating environment in which every move has a regulated consequence? At the spring event of IAPP KnowledgeNet Hungary, which I hosted, three speakers took up precisely this system-level question from three directions.
Three front lines
The pharmaceutical lifecycle. The panel’s pharmaceutical-industry speaker brought in the reality of global pharma: in a pharmaceutical company, AI is never a greenfield technology roll-out; it arrives in a lifecycle that is heavily regulated from the start, among research, manufacturing, quality management, pharmacovigilance, data protection, authorisation and documented evidence. The lesson: in a pharma setting, AI governance cannot be built around the tool alone; it has to be built into the product, data and evidence lifecycle.
The pharmacy front line. The panellist speaking for pharmacies looked at the same question from their side: the pharmacy is becoming less a dispensing point and more a hub of healthcare services, while patients already arrive informed by online sources and AI-based searches. According to a Hungarian survey of more than two hundred pharmacists (roughly a tenth of Hungary’s pharmacies), close to 70% of pharmacists consider it realistic that AI assistants will appear in pharmacy operations. Adoption, then, is pushed by developers and pulled by patients and care processes as well.
The data-asset paradox. The third panellist pointed to the dual nature of Hungary’s health data infrastructure: the EESZT (the national Electronic Health Service Space) and the central data assets are a foundation many countries can only dream of, yet having the data does not mean we have data-driven healthcare. The real question is what turns data into applications: working, clinically useful, regulated and sustainable solutions.
Where software stops being software
Read as a lawyer, the three perspectives converge on a single pattern, which at the event I called the regulatory sandwich effect. As soon as an AI system influences a patient pathway, triage, diagnostic pre-screening, clinical decision support or a therapeutic recommendation, the same system becomes an AI Act, MDR/IVDR, GDPR, NIS2, clinical evidence and liability question all at once. Every regime lands on it at the same time rather than in sequence, layer upon layer, as in a sandwich.
The patient, after all, meets a single working healthcare solution, not a set of separate regulatory regimes. That is why the real challenge is regulatory systems integration: bringing the legal, data protection, clinical, cybersecurity and product-law requirements and evidence together in one working governance model, organised around how the product actually works. A company that runs a separate programme for each regime does the same work several times over, and none of its programmes sees the gaps that open where the layers meet.
This is the healthcare version of what this journal argues in general: the rulebook is one operating layer, not five parallel folders. The life sciences sector is special only in that the layers here are more numerous and the stakes (patient safety, professional liability) higher. Teams that learn integration here will find other sectors straightforward by comparison.
This piece first appeared as a LinkedIn post in May 2026, following the IAPP KnowledgeNet Hungary chapter event “AI a laborban, a patikában és a betegágy mellett” (AI in the lab, the pharmacy and at the bedside); it is published here in an edited, self-contained form. The survey figures cited come from a Hungarian survey of pharmacists, as presented in the speakers’ public talks.