In January 2026, a landmark moment arrived quietly in the form of a joint document — but its implications are anything but quiet. The European Medicines Agency (EMA) and the U.S. Food and Drug Administration (FDA) jointly published ten guiding principles for good artificial intelligence (AI) practice across the entire medicines lifecycle. For researchers, clinicians, and developers working at the crossroads of AI and medicine, this is a development worth understanding deeply.
Why This Matters
Artificial intelligence is no longer on the horizon of medicine — it is already embedded within it. From predicting molecular drug candidates to monitoring post-market safety signals, AI tools are being used at virtually every stage of the pharmaceutical and clinical pipeline. Yet until recently, regulators in Europe and the United States had been developing their frameworks largely in parallel, creating a growing divergence that was proving to be a genuine barrier to innovation.
The new joint principles aim to dismantle that barrier. By establishing a shared language and shared expectations, EMA and FDA are signalling to the industry — and to the broader medical AI community — that accountability, safety, and transparency are not optional features of AI systems. They are requirements.
Olivér Várhelyi, the European Commissioner for Health and Animal Welfare, described the accord as “a first step of a renewed EU-US cooperation in the field of novel medical technologies,” with the goal of preserving a leading role in the global innovation race while ensuring the highest level of patient safety.
What the Principles Say
The ten guiding principles cover the full spectrum of AI use in drug development — from early discovery and clinical trials all the way through to manufacturing and post-market pharmacovigilance. Several themes stand out as especially significant.
Transparency and explainability. One of the central ambitions of the principles is to dismantle what industry experts call the “AI black box.” Pharmaceutical and biotechnology companies are expected to use plain language to explain the limitations of their AI systems and the nature of the underlying data. This is not simply a bureaucratic requirement — it reflects a genuine concern that patients, clinicians, and regulators must be able to understand and interrogate the outputs of AI systems that affect medical decisions.
Human oversight. The principles emphasise a human-centric approach throughout. AI must be applied in well-defined contexts, with clear mechanisms for human review, especially in high-stakes decisions about drug safety and efficacy.
Addressing “shadow use.” The joint document notably confronts the informal, undisclosed use of AI tools within pharmaceutical workflows — a practice that has raised concerns about data integrity and regulatory accountability. Making AI use visible and documented is a prerequisite for any meaningful oversight.
Privacy and data protection. Given that AI systems in medicine often draw on sensitive patient data, the principles reinforce the obligation to protect personal information in line with existing legal frameworks.
The European Regulatory Landscape in 2026
The EMA/FDA accord does not exist in a vacuum. It arrives at a moment when the EU’s broader AI regulatory architecture is coming into force in a very tangible way.
The EU AI Act — the world’s first comprehensive AI law — entered into force in August 2024. Its most demanding requirements, those governing so-called “high-risk” AI systems, take effect on 2 August 2026. Medical AI systems that influence clinical decisions are firmly within the high-risk category, meaning that companies and institutions deploying such tools must meet strict requirements around risk management, data quality, transparency, and human oversight.
At the same time, the European Commission published a proposal in December 2025 to harmonise the AI Act’s requirements for medical devices with the existing EU Medical Device Regulation (MDR) and In Vitro Diagnostics Regulation (IVDR). If adopted — and observers anticipate this could happen by summer 2026 — it would mean that conformity assessments for AI-enabled medical devices are managed entirely under the MDR framework, reducing the duplicative compliance burden that has worried manufacturers and innovators alike.
For hospitals, research institutions, and health technology developers across the EU — including in Croatia — this convergence of regulatory frameworks is creating both clarity and urgency.
What It Means for Medical AI Practitioners
For those actively working with AI in clinical and research settings, several practical consequences follow from these developments.
First, documentation is now a strategic priority. AI systems used in any part of the medicines lifecycle — including clinical decision support, imaging analysis, or trial data processing — need to be thoroughly documented, with clear records of how models were trained, validated, and deployed.
Second, regulatory literacy is becoming a clinical competency. Understanding the distinction between a high-risk and a non-high-risk AI system under the EU AI Act, or knowing when a tool requires CE marking under the MDR, is no longer the sole province of legal teams. Clinicians and data scientists who build or use these tools need to be conversant in the regulatory landscape.
Third, the EMA/FDA principles offer a practical reference framework. Even for institutions not directly involved in drug development, the ten principles provide useful guidance on what responsible AI use in medicine looks like: well-defined context of use, explainability, human oversight, and rigorous data governance.
A Moment of Alignment
The significance of the EMA/FDA accord goes beyond its immediate regulatory scope. The European Federation of Pharmaceutical Industries and Associations (EFPIA) noted that the principles “help create a more coherent environment for scaling AI tools globally and for engaging with regulators in a consistent manner.” When the two largest pharmaceutical regulatory bodies in the world agree on foundational principles, it shapes expectations everywhere — including for academic medical centres, hospital systems, and research consortia developing AI tools that may never be submitted to a regulatory body directly, but whose outputs feed into the broader evidence ecosystem.
For the AI2MED project community, this is an encouraging signal. The regulatory conversation is maturing. The question is no longer whether AI belongs in medicine, but how to deploy it responsibly, transparently, and equitably. The EMA/FDA principles, imperfect and high-level as they are, represent a meaningful step in the right direction.
This article draws on the joint EMA/FDA guiding principles published on 14 January 2026, reporting by Health Policy Watch, and the ongoing regulatory analysis of the EU AI Act’s application to the medical sector. For the full text of the guiding principles, visit the EMA website.
