Introduction to the Basic Principles for AI Use in Regulatory Activities
On November 2, 2026, Pharmaceuticals and Medical Devices Agency (PMDA) released a new Early Consideration titled Basic Principles for the Use of AI in Regulatory Activities. The document provides guiding principles for the responsible use and development of artificial intelligence, including generative AI, in activities involving regulatory judgment and decision-making.
The guidance recognizes the potential of AI to improve the efficiency and quality of medical product development and evaluation. At the same time, PMDA emphasizes that AI should ultimately contribute to protecting the interests of patients and the public.
As an Early Consideration, the document does not establish new regulatory requirements. Instead, it represents principles intended to promote responsible AI use while managing potential risks. It may be updated as AI technologies, practical experience, and international regulatory thinking continue to evolve.
Broad Application Across the Medical Product Life Cycle
The principles apply to the use of AI in regulatory activities throughout the entire product life cycle, from development through manufacturing and post-marketing activities.
The scope includes:
The principles cover pharmaceuticals, quasi-drugs, cosmetics, medical devices, and regenerative medical products. They are intended for a broad range of stakeholders, including PMDA, applicants, marketing authorization holders, manufacturers, sponsors, CROs, and developers and providers of AI systems.
Human-Centric Principle: AI Supports, Rather Than Replaces, Human Judgment
One of the most important messages is the Human-Centric Principle.
PMDA states that AI is intended to support, rather than replace, human judgment and decision-making. The interests of patients and the public should remain the highest priority, and ultimate responsibility for decisions involving AI remains with humans.
This principle is particularly important in regulatory activities, where AI outputs may influence decisions affecting product quality, safety, efficacy, or patient access.
Risk-Based Approach to AI Use
The guidance adopts a risk-based approach to AI governance.
The level of control and risk-management measures applied to an AI system should reflect the probability and severity of undesirable events that could arise from its use.
This suggests that AI applications should not necessarily be governed identically. An AI tool used for a relatively low-impact administrative activity may warrant a different level of control from an AI system directly supporting a regulatory judgment or other consequential decision.
Clearly Define the Context of Use
Before AI is used or developed, its context of use should be clearly defined.
This includes consideration of:
Defining these parameters provides the foundation for determining the appropriate level of validation, oversight, explainability, and risk management.
Data Governance and Traceability
PMDA places significant emphasis on data governance.
Organizations should establish appropriate governance and management processes to maintain a reliable data foundation while protecting privacy. Data entered into or processed by AI should be appropriate for the intended use.
For data used to develop AI, organizations should appropriately document and manage:
These principles are particularly relevant to regulated environments where reproducibility, data integrity, and auditability are critical.
Fairness, Accountability, and Transparency
The guidance emphasizes the need to minimize inappropriate bias in AI outputs and in decisions influenced by those outputs.
Stakeholders should provide clear and understandable information regarding:
Information should also flow appropriately between AI developers, AI users, and relevant stakeholders, which may include patients, the public, and healthcare professionals.
Multidisciplinary Expertise
AI governance should not rely solely on technical expertise.
PMDA recommends bringing together multidisciplinary perspectives covering:
Patient and public perspectives should also be incorporated when appropriate.
This suggests that effective AI governance in pharmaceutical companies will require collaboration across Regulatory Affairs, Clinical Development, Safety, Quality, IT/Data Science, Legal, Compliance, and other functions rather than ownership by an AI or IT team alone.
Validation Before Use
Before AI is implemented, its performance should be appropriately evaluated and validated based on its intended use, scope, and associated risks.
Importantly, evaluation should address not only the technical performance of the AI model itself but also how the AI system performs in its actual operating environment.
The AI model or system should also be reliable and robust, with a level of explainability appropriate to its intended use and risk.
Life Cycle Management of AI Systems
AI validation should not be viewed as a one-time activity.
Following implementation, organizations should continuously monitor and evaluate the performance of AI models and systems. Periodic reviews should be performed to maintain and improve their suitability over time.
This life cycle management concept is particularly important because AI technologies, models, data, and operating environments can change after initial deployment.
Human Oversight and AI Literacy
The guidance specifically warns against excessive reliance on AI outputs.
Organizations should establish appropriate human oversight based on the characteristics, limitations, intended use, and risk of the AI system. Human review and judgment should be incorporated where appropriate, with operational processes that allow humans to intervene.
At the same time, effective oversight requires AI literacy. Users should continuously improve their understanding of AI capabilities and limitations and remain informed about emerging technologies and newly identified risks.
Compliance with Existing Laws, Regulations, and Standards
Use of AI does not replace or override existing regulatory responsibilities.
Organizations remain responsible for complying with applicable laws and regulations and following relevant ethical, technical, scientific, cybersecurity, and regulatory standards or norms.
The Early Consideration therefore provides an additional governance framework for AI rather than creating an alternative regulatory pathway.
Conclusion
The PMDA Early Consideration represents an important step toward establishing a practical regulatory framework for AI-enabled drug development and regulatory activities in Japan.
Rather than restricting the use of AI, PMDA establishes a set of principles for responsible adoption: human accountability, risk-based governance, clearly defined context of use, strong data governance, transparency, multidisciplinary oversight, validation, continuous life cycle management, and AI literacy.
For pharmaceutical companies, the practical message is clear: AI can play an increasingly important role in regulatory activities, but its use should be supported by governance and controls proportionate to the potential impact of its outputs on regulatory judgments and decisions.
Reference
Early Consideration | Pharmaceuticals and Medical Devices Agency