The Ministry of Food and Drug Safety (MFDS) is preparing to implement its “AI Reviewer” system, aimed at enhancing the efficiency and accuracy of drug evaluations. The initiative is expected to launch as early as next year, beginning with chemical drug reviews, following the completion of AI model development for report drafting and impurity assessments.

Since early this year, the MFDS has been conducting a project titled “Research on Regulatory Strategies for AI-Assisted Drug Evaluation.”

At a recent press briefing, Ji-Won Jung, Director of the Medical Products Research Division, stated, “When companies submit approval applications for new or generic drugs, MFDS reviewers prepare evaluation reports that may include recommendations for approval or requests for supplementary data. This process often involves analyzing thousands to tens of thousands of pages, requiring significant time and resources.”

Jung further explained that integrating AI could help accelerate the review process. “With AI support, reviewers could receive summarized reports and visualized data, enabling faster assessments,” she said. “The goal of this study is to determine whether AI can assist in drafting review reports and what the optimal format would be.”

A key focus of the project is the development of an AI model for impurity evaluation. Sang-ae Park, Head of Drug Standards at the Drug Evaluation Department of the National Institute of Food and Drug Safety Evaluation, noted, “Impurity assessment heavily relies on historical data. When a company submits documentation, the first step is to determine whether the impurity has been previously reviewed and how MFDS handled it at the time.”

“If no precedent exists, a more thorough review is required,” she continued. “However, if there is historical alignment, past decisions can inform the current evaluation. This process demands extensive data analysis, which we aim to streamline through AI.”

AI experts are optimistic about the potential of the “Impurity AI Reviewer.” One specialist in AI-driven drug development commented, “With the increasing emergence of new impurities, now is the time to structure MFDS’s historical impurity evaluations—especially nitrosamines like those found in valsartan—into a big data format. Fine-tuning AI models with this data may allow for greater precision than human reviewers.”

He added, “Defining impurity categories and establishing conditional approval criteria is essential. If these are clearly set, the AI model could potentially issue suitability or deficiency decisions rapidly, improving review speed.”

MFDS data highlights the urgency of such innovation. According to its report “Major Deficiency Cases in Generic Drug Quality Reviews” covering January 2023 to June 2024, impurity-related issues were the top cause of additional documentation requests in active pharmaceutical ingredients (22.2%), with genotoxic impurities making up 37.4% of those cases.

Park emphasized, “We are working on structuring and accumulating summary reports and impurity data while validating the accuracy of AI-generated outputs. Although data collection takes time, once the system is fully developed, it will significantly improve the review process.”

Jung concluded, “This is not a one-time initiative. Just as the U.S. FDA has adopted AI review tools, the MFDS is committed to establishing its own AI-based review infrastructure. Upon completion of internal evaluations, we plan to introduce the AI Reviewer system for chemical drugs as early as next year.”