# Federating AI-related regulations for human therapeutics: an AI-enabled, continuously updating regulatory intelligence system

## TL;DR

This paper introduces AICURIS, an AI system that tracks and compares global AI-related healthcare regulations to reduce disparities and improve access to AI-based therapies.

## Contribution

The novel contribution is an AI-enabled regulatory intelligence system that continuously monitors and compares AI-related regulations across jurisdictions.

## Key findings

- An AI model achieved 95% recall for identifying AI-related regulatory content from over 400,000 documents.
- The system enables structured comparison of regulations across the FDA, EMA, and WHO.
- The system supports real-time monitoring and identifies alignment and divergence in regulatory approaches.

## Abstract

Global regulation of artificial intelligence (AI) in healthcare remains highly fragmented. Over 1,000 national AI-related regulations and policy frameworks have been introduced across more than 70 countries, most originating from high-income nations. While major drug regulatory authorities, including those in the United States and European Union, and others like WHO, have issued guidance on AI-enabled systems, platforms, and products, no unified source exists that compiles or compares how these bodies govern the broader AI-enabled therapeutic ecosystem. This lack of alignment contributes to disparities in interpretation and implementation, potentially delaying global access to novel AI-based therapies. To address this gap, we developed AI-enabled, continuously updated regulatory intelligence system (AICURIS), a comprehensive AI-enabled regulatory intelligence system trained initially on AI-related regulations from sentinel regulatory authorities (e.g., FDA, EMA, WHO), designed to continuously monitor and classify AI-relevant regulatory content and to enable structured comparison, identification of alignment and divergence, and evidence-informed discussion on regulatory convergence as its coverage expands across jurisdictions. Using over 400,000 regulatory documents published since 2019, an AI-enabled hybrid semantic similarity and keyword scoring model achieved 95% recall for high-confidence AI-related content. The following results summarize the key findings from the analysis of AI-related regulatory documents across three major agencies (United States Food and Drug Administration (FDA), European Medicines Agency (EMA), and the World Health Organization (WHO)). The findings are presented in six parts: (1) document corpus composition and filtering outcomes, (2) model optimization, (3) classification model performance, (4) classification ensemble model validation, (5) cross-agency bias mitigation, and (6) real-time monitoring capabilities. This study demonstrates that AI-powered, data-driven approaches can effectively federate AI-related drug regulations across jurisdictions, reducing global disparities, and enabling more equitable, collaborative, and efficient human therapeutics innovation.