Current challenges and the way forward for regulatory databases of artificial intelligence as a medical device
npj Digital Medicine, 2026
Strong regulatory oversight depends on strong regulatory databases, yet today’s databases are often not fit for purpose for AI as a medical device (AIaMD). Many stakeholders (regulators, clinicians, procurement teams, patients, researchers, and manufacturers) rely on these systems to understand what’s approved, what evidence supports it, and what safety signals are emerging, but the data needed to make safe, informed decisions is frequently hard to find, incomplete, or fragmented.
The authors identify four core problems: limited search functionality, no consistent way to identify AI-enabled devices because AI-specific terminology is missing from common nomenclature systems, limited transparency around clinical evaluation and real-world performance, and fragmentation across multiple databases for approvals, vigilance, adverse events, and recalls, making lifecycle tracking burdensome and inconsistent across jurisdictions. Researchers try to work around these gaps using “snowball” searches, literature cross-checking, and automated/NLP approaches, efforts that still can’t fully overcome the underlying lack of accessible, standardised data.
Looking forward, the paper proposes a roadmap toward a next-generation regulatory data ecosystem. Firstly, it is suggested tp create fit-for-purpose local databases with better UX, AI-specific tagging, machine-readable formats, and a minimum public dataset. Secondly, to enable federated analysis across jurisdictions to improve traceability while preserving sovereignty. Finally, it is proposed to evolve into a “learning” ecosystem where post-market data helps detect safety signals and performance drift more proactively.
The article below is a preprint of the manuscript.