How AI is reshaping EU medicines regulation: Key findings from the 2024 AI Observatory report

The European Medicines Regulatory Network (EMRN) has published its first AI Observatory report, providing a comprehensive overview of how AI is being adopted across the EU regulatory system. The report highlights AI’s growing role in medicines development and evaluation, covering data insights, productivity tools, scientific advice, and advanced modelling techniques.

Published jointly by the European Medicines Agency (EMA) and Heads of Medicines Agencies (HMA), the report marks a significant step in implementing the EMA-HMA multi-annual AI workplan. It was developed under the leadership of the Network Data Steering Group (NDSG) to consolidate experience across the network and identify strategic directions for future AI integration.

The Observatory was established to provide continuous oversight of emerging AI technologies and their regulatory implications. It aims to collect, assess, and share real-world examples of AI use across the EMRN, helping to inform policy and support robust decision-making. As noted in the report, “AI requires constant monitoring both for the emergence of novel systems and [for] impact.”

AI across the medicine lifecycle

The 2024 report shows that AI is increasingly being applied throughout the medicine lifecycle, although most regulatory interactions currently occur at the pre-authorization stage. In 2024, the EMA was involved in seven qualification procedures and four scientific advice processes related to AI tools. These included deep learning-based image analysis for liver disease, machine learning to assess radiographic progression in psoriatic arthritis, and digital twins in rare disease trials.

AI use cases were explored across multiple regulatory forums, including Scientific Advice, Innovation Task Force (ITF), and Portfolio and Technology meetings. Therapeutic areas covered ranged from oncology and neurology to dermatology and immuno-inflammation. Applications included automated histological scoring, endpoint adjudication, treatment response prediction, and patient selection for clinical trials using real-world data (RWD) and synthetic control arms.

The report also captured feedback from developers, highlighting AI’s expanding role in early drug development, manufacturing, clinical trial optimization, and post-authorization safety monitoring. Developers expressed a need for regulatory clarity, with some calling for a more globally aligned and adaptive framework:

“We encourage EMA to further develop its thinking in this area… there is increasing interest in how to use AI/ML as a lever to automate, e.g., data capture/verification, monitoring… and potential for developing algorithms for diagnosis and predictive models… Key principles in these areas would be useful.”

Others emphasized the importance of flexibility in regulatory guidance:

“A nimble approach from EMA to fill gaps in existing regulatory recommendations [would] encourage further investment in the use of AI/ML across drug development.”

AI for automation and productivity

Beyond its scientific applications, AI is being used to automate regulatory workflows and improve productivity across EU agencies. In 2024, the network reported 27 AI tools in use, 14 of which originated from national agencies. At EMA, operational tools in production included systems for anonymization, referral template editing, invoice processing, and redaction of commercial information. Emerging pilots, such as ChatGPT@EMA and speech-to-text services, are also being tested to support internal workflows.

The EMA’s Scientific Explorer, launched in March 2024, is an AI-powered knowledge mining platform that enables regulators to search and analyze prior regulatory decisions, particularly scientific advice letters, for both human and veterinary medicines.

At the national level, several innovative tools have been deployed. The Swedish Medical Products Agency's M-RECON and PICROSS enable semantic document matching and cross-referencing of product information, while Denmark’s PHarmacovigilance by AI Real-time (PHAIR) system integrates national health data to monitor vaccine safety in real-time. The Netherlands’ MEB is piloting natural language processing tools to cluster uncertainties in regulatory documents.

The EMA Health Data Lab also piloted several AI-driven tools in 2024, including EurEKA for extracting adverse reaction data, AERGIA for automated signal adjudication, and ERATO, which screens scientific literature for emerging safety concerns using generative AI.

Horizon scanning and strategic foresight

To complement its inventory of experience, the AI Observatory also conducted a horizon scanning exercise based on peer-reviewed publications, preprints, and EU-funded R&D projects. This review helped identify areas where additional guidance, collaboration, or infrastructure may be needed to support the integration of AI across regulatory activities.

Notably, the publication of the EU AI Act in 2024 provides a broader governance context. While developers acknowledged its importance, they also raised questions about its application to pharmaceutical R&D – particularly the distinction between regulated and exploratory uses of AI. As one developer noted:

“We believe that the EU AI Act R&D exemption described in Recital 25, Articles 2.6 and 2.8… should apply to Pharma R&D processes if treatment of patients is not directly affected… How does EMA define and assess the relative risk of AI… for different uses in the medicine lifecycle, and the framework being used for such assessments?”

The AI Observatory’s findings will feed into the Data and AI in medicines regulation workplan 2025–2028, and into broader strategic efforts to modernize regulatory science. As noted in the report,

“Learnings from the experience of compiling this first EMRN AI Observatory report will be considered by the NDSG to improve future reports.”