Reflection paper on the use of Artificial Intelligence (AI) in the medicinal product lifecycle

Context and history

EMA's reflection paper sets out current thinking on AI and machine learning across a medicine's lifecycle. The draft was published on 19 July 2023, with consultation closing on 31 December 2023. It was adopted by CHMP on 9 September 2024 and by CVMP on 11 September, and published on 30 September 2024. It is a reflection paper, so it is not a binding rule.

Who and what it applies to

Applicants and marketing authorisation holders for human and veterinary medicines using AI at any stage.

What it contains

Introduction and definitions

What is meant by AI and machine learning.

General considerations

A risk-based approach, and responsibility of the applicant or holder.

Lifecycle stages

Drug discovery, non-clinical development, clinical trials, precision medicine, product information, manufacturing and post-authorisation.

Regulatory interactions

Early engagement.

Technical aspects

Data acquisition, training, validation and test datasets, model development, performance, interpretability and deployment.

Governance, integrity and ethics

Governance, data protection and trustworthy AI.

Key points

• It is the responsibility of the sponsor, applicant, holder or manufacturer to ensure algorithms, models, datasets and pipelines are fit for purpose.
• EMA notes these standards may be stricter than standard data science practice.
• A risk-based approach applies across the lifecycle.
• For manufacturing, quality risk management applies, taking patient safety, data integrity and product quality into account.
• For human medicines, EMA says the principles of ICH Q8, Q9 and Q10 should be considered, pending revision of GMP standards.
• Early dialogue with regulators is encouraged.

In practice

General reading, not advice. This is drafted with AI and may not fit your situation. It does not replace the official text, your own quality system or qualified advice, and an inspector may take a different view.

• Read this with draft Annex 22 for manufacturing uses, because the paper predates it.
• Do not rely on general data science practice as your standard of evidence.
• Keep enough technical detail to explain a model to a regulator.

Read from the source document on 1 Nov 2026. This is our own description, not the official text. Check the full dated history in the change log below.