Hello, this is JNPMEDI.
The use of AI in drug development is no longer limited to experimental efforts within a small number of companies or research organizations. From candidate discovery and optimization to clinical design and data analysisโand further into manufacturing processes and post-marketing safetyโAI is already being applied across the entire drug development lifecycle. What was once used selectively in early-stage research has now become a factor that generates evidence or influences analysis throughout the full product lifecycle, including nonclinical, clinical, post-marketing, and manufacturing stages.
Amid these changes, the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) jointly released the โGuiding Principles of Good AI Practice in Drug Developmentโย in January 2026. Rather than introducing new regulatory requirements, this document serves as a framework that organizes how AIโalready in use in practiceโshould be viewed and managed from a regulatory perspective. As a result, the key regulatory question is shifting.
It is no longer โCan AI be used?โ but โHow can development and operational outcomes involving AI be explained and reviewed from a regulatory standpoint?โ
AI Use Is No Longer a Technical IssueโIt Has Become a Structural One
Recent developments in the global pharmaceutical industry show that AI is evolving beyond individual algorithms or analytical tools into platform- and infrastructure-level systems.
At the 2026 J.P. Morgan Healthcare Conference, Eli Lilly and NVIDIA announced the establishment of an โAI Co-Innovation Labโ, outlining plans to build an AI-driven drug development environment based on large-scale research data. This collaboration focuses on building an AI utilization environment grounded in large-scale data and research infrastructure, rather than on the performance of individual AI models.
This trend has important implications from a regulatory perspective. Rather than focusing on AI performance itself, regulators are increasingly concerned with what data is used, how that data is managed and controlled, and how AI-generated results are applied within decision-making processes across the broader operational structure.
Similar changes are also emerging in Korea, particularly in the manufacturing domain.
Samsung Biologics has introduced an operational framework that integrates production data across processes, leveraging digital twin technology (which replicates real-world processes in a virtual environment for simulation and monitoring) and data lake architecture (a centralized repository that stores structured and unstructured data in its original form). This demonstrates that AI is no longer confined to research stages but is becoming part of GMP environments and operational management systems.
In addition, Korean pharmaceutical companies such as Yuhan Corporation and Dong-A ST are using AI-based predictive models to estimate the efficacy and pharmacokinetics of candidate compounds, while improving R&D efficiency.
These examples show that the discussion is no longer about whether to adopt AI, but about how to use it within the broader context of organizational structures, data, and operational systems.
Key Takeaways from the FDAโEMA Guiding Principles
(Source: FDA)
At first glance, the ten principles presented by the FDA and EMA may appear broad. However, when viewed from a regulatory perspective, several consistent directions become clear.
1. Not โWhat Can AI Do?โ but โWho Uses It, and Under What Accountability Structure?โ
- Human-centric by design
- Risk-based approach
- Adherence to standards
- Clear context of use
- Multidisciplinary expertise
These principles assume that AI is not an independent decision-making entity, but a system that supports and extends human judgment within a defined structure of responsibility and accountability.
2. Not the โOutcome,โ but the โTraceability of the Processโ
The following principles show that regulators focus less on the outputs generated by AI and more on how those outputs are produced and managed:
- Data governance and documentation
- Model design and development practices
- Risk-based performance assessment
The requirement that data sources, processing steps, and model development and validation processes be traceable and explainable reflects a clear position: AI should not be treated as a โblack box.โ
3. AI Is Not โCompleted After DevelopmentโโIt Requires Ongoing Management
The following principles indicate that AI is not a one-time development output, but a system that requires continuous monitoring and management:
- Life cycle management
- Clear, essential information
This includes responsibilities related to data drift (changes in input data distribution over time that may affect model performance), performance changes, update history, and clear communication of limitations.
This reflects the view that AI should be understood not as a fixed product, but as an operational system subject to ongoing management.
The Focus of Regulation Is Not Performance, but Explainability
The core message of these ten principles is clear. Regulators do not evaluate AI solely based on how accurate it is. Instead, they focus on what data and assumptions it is built upon, and how it is managed within a defined operational framework.
When AI is used in clinical design, regulatory decision-making, or manufacturing operations, its outputs no longer remain as technical references. They become evidence directly used in regulatory evaluation.
Accordingly, the FDA and EMA are less concerned with the novelty or performance of algorithms themselves, and more focused on how AI-involved decisions can be explained and how accountability for those decisions is managed.
These guiding principles are not intended to restrict the use of AI, but rather to organize AIโalready in useโinto a framework that can be managed from a regulatory perspective. As noted in the KoBIA Brief, these principles are likely to serve as a starting point for international regulatory harmonization, standard development, and the establishment of national AI-related guidelines. This suggests that as the use of AI expands, regulatory frameworks will become more detailed over time.
Implications for Practice and Industry Structure
These principles do not immediately create new regulatory obligations. However, from the perspective of clinical, regulatory, and data operations, their mid- to long-term impact is clear.
The key question going forward is no longer โWas AI used?โ but โWithin what management framework was that AI used?โ In other words, the ability to explain the entire process and governance structure behind AI-generated outcomes will become just as important as the outcomes themselves.
Preparing for the Post-Adoption Phase of AI
Now that AI-driven drug development has become routine, regulatory discussions are no longer centered on technological performance. Instead, the focus has shifted to the role AI plays across the clinical and regulatory lifecycle, and the accountability structures under which it operates. The FDAโEMA guiding principles are less a set of restrictions and more a baseline for organizing AIโalready in useโinto a manageable framework across the clinical and regulatory lifecycle.
Competition in AI-driven drug development is also moving beyond pure technological capability toward the ability to integrate AI with clinical and regulatory strategy.
If you would like to explore how AI utilization can be aligned with clinical and regulatory execution, we invite you to connect with JNPMEDI.
๐โโ๏ธ Contact JNPMEDI for inquiries
โข Reference: [KoBIA Brief] EMA-FDA ์์ฝํ์ ์ฃผ๊ธฐ AI ํ์ฉ์ ๋ํ 10๊ฐ์ง ์์น ์๋ฆฝ
Hello, this is JNPMEDI.
The use of AI in drug development is no longer limited to experimental efforts within a small number of companies or research organizations. From candidate discovery and optimization to clinical design and data analysisโand further into manufacturing processes and post-marketing safetyโAI is already being applied across the entire drug development lifecycle. What was once used selectively in early-stage research has now become a factor that generates evidence or influences analysis throughout the full product lifecycle, including nonclinical, clinical, post-marketing, and manufacturing stages.
Amid these changes, the U.S. Food and Drug Administration (FDA) and the European Medicines Agency (EMA) jointly released the โGuiding Principles of Good AI Practice in Drug Developmentโย in January 2026. Rather than introducing new regulatory requirements, this document serves as a framework that organizes how AIโalready in use in practiceโshould be viewed and managed from a regulatory perspective. As a result, the key regulatory question is shifting.
It is no longer โCan AI be used?โ but โHow can development and operational outcomes involving AI be explained and reviewed from a regulatory standpoint?โ
AI Use Is No Longer a Technical IssueโIt Has Become a Structural One
Recent developments in the global pharmaceutical industry show that AI is evolving beyond individual algorithms or analytical tools into platform- and infrastructure-level systems.
At the 2026 J.P. Morgan Healthcare Conference, Eli Lilly and NVIDIA announced the establishment of an โAI Co-Innovation Labโ, outlining plans to build an AI-driven drug development environment based on large-scale research data. This collaboration focuses on building an AI utilization environment grounded in large-scale data and research infrastructure, rather than on the performance of individual AI models.
This trend has important implications from a regulatory perspective. Rather than focusing on AI performance itself, regulators are increasingly concerned with what data is used, how that data is managed and controlled, and how AI-generated results are applied within decision-making processes across the broader operational structure.
Similar changes are also emerging in Korea, particularly in the manufacturing domain.
Samsung Biologics has introduced an operational framework that integrates production data across processes, leveraging digital twin technology (which replicates real-world processes in a virtual environment for simulation and monitoring) and data lake architecture (a centralized repository that stores structured and unstructured data in its original form). This demonstrates that AI is no longer confined to research stages but is becoming part of GMP environments and operational management systems.
In addition, Korean pharmaceutical companies such as Yuhan Corporation and Dong-A ST are using AI-based predictive models to estimate the efficacy and pharmacokinetics of candidate compounds, while improving R&D efficiency.
These examples show that the discussion is no longer about whether to adopt AI, but about how to use it within the broader context of organizational structures, data, and operational systems.
Key Takeaways from the FDAโEMA Guiding Principles
(Source: FDA)
At first glance, the ten principles presented by the FDA and EMA may appear broad. However, when viewed from a regulatory perspective, several consistent directions become clear.
1. Not โWhat Can AI Do?โ but โWho Uses It, and Under What Accountability Structure?โ
These principles assume that AI is not an independent decision-making entity, but a system that supports and extends human judgment within a defined structure of responsibility and accountability.
2. Not the โOutcome,โ but the โTraceability of the Processโ
The following principles show that regulators focus less on the outputs generated by AI and more on how those outputs are produced and managed:
The requirement that data sources, processing steps, and model development and validation processes be traceable and explainable reflects a clear position: AI should not be treated as a โblack box.โ
3. AI Is Not โCompleted After DevelopmentโโIt Requires Ongoing Management
The following principles indicate that AI is not a one-time development output, but a system that requires continuous monitoring and management:
This includes responsibilities related to data drift (changes in input data distribution over time that may affect model performance), performance changes, update history, and clear communication of limitations.
This reflects the view that AI should be understood not as a fixed product, but as an operational system subject to ongoing management.
The Focus of Regulation Is Not Performance, but Explainability
The core message of these ten principles is clear. Regulators do not evaluate AI solely based on how accurate it is. Instead, they focus on what data and assumptions it is built upon, and how it is managed within a defined operational framework.
When AI is used in clinical design, regulatory decision-making, or manufacturing operations, its outputs no longer remain as technical references. They become evidence directly used in regulatory evaluation.
Accordingly, the FDA and EMA are less concerned with the novelty or performance of algorithms themselves, and more focused on how AI-involved decisions can be explained and how accountability for those decisions is managed.
These guiding principles are not intended to restrict the use of AI, but rather to organize AIโalready in useโinto a framework that can be managed from a regulatory perspective. As noted in the KoBIA Brief, these principles are likely to serve as a starting point for international regulatory harmonization, standard development, and the establishment of national AI-related guidelines. This suggests that as the use of AI expands, regulatory frameworks will become more detailed over time.
Implications for Practice and Industry Structure
These principles do not immediately create new regulatory obligations. However, from the perspective of clinical, regulatory, and data operations, their mid- to long-term impact is clear.
The key question going forward is no longer โWas AI used?โ but โWithin what management framework was that AI used?โ In other words, the ability to explain the entire process and governance structure behind AI-generated outcomes will become just as important as the outcomes themselves.
Preparing for the Post-Adoption Phase of AI
Now that AI-driven drug development has become routine, regulatory discussions are no longer centered on technological performance. Instead, the focus has shifted to the role AI plays across the clinical and regulatory lifecycle, and the accountability structures under which it operates. The FDAโEMA guiding principles are less a set of restrictions and more a baseline for organizing AIโalready in useโinto a manageable framework across the clinical and regulatory lifecycle.
Competition in AI-driven drug development is also moving beyond pure technological capability toward the ability to integrate AI with clinical and regulatory strategy.
If you would like to explore how AI utilization can be aligned with clinical and regulatory execution, we invite you to connect with JNPMEDI.
๐โโ๏ธ Contact JNPMEDI for inquiries
โข Reference: [KoBIA Brief] EMA-FDA ์์ฝํ์ ์ฃผ๊ธฐ AI ํ์ฉ์ ๋ํ 10๊ฐ์ง ์์น ์๋ฆฝ