FDA and EMA Align on Good AI Practice in Pharma

The FDA and the European Medicines Agency (EMA) have agreed on a shared set of guiding principles to support the safe, ethical and reliable use of artificial intelligence (AI) across the full lifecycle of medicines, from early research and clinical trials to manufacturing and post-marketing safety monitoring.

According to the FDA, its Center for Drug Evaluation and Research (CDER) and Center for Biologics Evaluation and Research (CBER) worked with EMA to develop ten principles that industry and product developers can consider when using AI to generate evidence for drugs and biological products.

The agencies said the goal is to ensure that AI-enabled tools support regulatory decision-making while maintaining patient safety, data integrity and established standards for quality, safety and efficacy.

In the joint framework, AI is defined as system-level technologies used to generate or analyze evidence across nonclinical, clinical, manufacturing and post-approval phases.

Regulators note that AI may help accelerate development timelines, strengthen pharmacovigilance and improve prediction of toxicity and efficacy in humans. They note AI could also support efforts to reduce reliance on animal testing, provided such tools are appropriately developed and validated.

Here is a closer look at the 10 guiding principles for good AI practice in drug development:

Human-Centric and Ethical by Design

A central theme is that AI systems should be human-centric and aligned with ethical values, meaning that people remain responsible for decisions and that meaningful human oversight is maintained throughout development and use.

Risk-Based Approach

The principles take a risk-based approach, calling for validation, performance testing, risk mitigation and regulatory oversight that are proportionate to a system’s intended role and its potential impact on patients or product quality.

Clear Context of Use

Each application should also have a clearly defined context of use, specifying what decisions or processes the AI is meant to support and the setting in which it is intended to be applied.

Multidisciplinary Expertise

The agencies further stress the importance of multidisciplinary expertise, integrating knowledge of AI methods with understanding of the relevant scientific, clinical, manufacturing and regulatory settings.

Data Governance and Documentation

Robust data governance is another core requirement. Developers are expected to document data sources, processing steps and analytical decisions in a detailed, traceable and verifiable manner, consistent with Good Practice (GxP) expectations, while maintaining appropriate privacy and security protections throughout the technology’s lifecycle.

Model Design and Development Practices

From a technical perspective, the framework calls for established model and system design and software engineering practices. AI tools should be developed using data that are fit for purpose and assessed for transparency, robustness, generalizability and where relevant, interpretability or explainability, so that their outputs can be understood, trusted and relied upon for regulatory and clinical decision-making.

Risk-Based Performance Assessment

Performance assessment should also follow a risk-based approach and evaluate the complete system, including interactions between humans and AI. Validation and testing methods should rely on appropriate datasets and metrics aligned with the intended context of use.

Lifecycle Management

Lifecycle management is emphasized as well. The principles call for quality management systems to support ongoing monitoring and periodic re-evaluation of AI technologies after deployment, including processes to detect and address issues such as data drift or declining performance over time.

Clear, Essential Information

Clear communication forms the final pillar. Plain-language information should be provided to relevant audiences, including regulators, users and, where appropriate, patients, describing an AI system’s purpose, performance, limitations, underlying data, updates and the extent to which its outputs can be interpreted.

Both agencies emphasized that the ten principles are intended as a foundational framework rather than prescriptive technical rules, and that they will be supplemented by more detailed, jurisdiction-specific guidance as legal and regulatory requirements evolve and experience with AI in drug development grows.

How AI Is Being Applied in Drug Development

AI is increasingly being embedded into specific stages of drug research and development.

In January, Converge Bio and Proxima (formerly VantAI) raised $25 million and $80 million, respectively, to expand AI platforms used for early drug discovery, including identifying new targets, designing antibody and small-molecule candidates and modelling protein-protein interactions underlying modalities such as molecular glues and PROTACs.

At the data and infrastructure level, Lilly and NVIDIA announced a joint AI co-innovation lab to develop foundation models, robotics-enabled experimental workflows and digital twins to support continuous, AI-assisted research across discovery, clinical development and manufacturing. Separately, AstraZeneca acquired Modella AI to integrate multimodal foundation models and AI agents into its oncology pipeline, supporting biomarker discovery and data-driven clinical development.

If you want your company to be featured on Xtalks.com, please email [email protected].