Regulating AI in drug discovery: EMA FDA ICH
Regulating AI in drug discovery is no longer a US-only briefing note for EU and UK sponsors. EMA’s adopted AI reflection paper, joint FDA–EMA good-AI principles, ICH M15 on model-informed development, and MHRA’s AI strategy now set concrete expectations for how AI/ML evidence must be governed, documented, and defended in marketing dossiers.
Contents11 sections
Key Takeaways
- EMA adopted its AI reflection paper (EMA/CHMP/CVMP/83833/2023) on 9 September 2024; it spans discovery to pharmacovigilance.
- FDA and EMA jointly issued 10 Guiding Principles of Good AI Practice in Drug Development in January 2026.
- ICH M15 (Step 5 via EMA) treats AI/ML as a modelling method under model-informed drug development (MIDD) assessment.
- MHRA’s April 2024 AI strategy maps five UK White Paper principles onto medicines and AI as a medical device (AIaMD).
AI drug-discovery regulation at a glance
| Instrument | Issuer | Status / date | EU/UK sponsor focus |
|---|---|---|---|
| AI in medicinal product lifecycle | EMA (CHMP/CVMP) | Adopted 9 Sep 2024 | Risk-based AI/ML across R&D to post-auth |
| Good AI Practice principles (10) | FDA + EMA | Published Jan 2026 | Human-centric, COU, life-cycle controls |
| AI credibility draft guidance | FDA | Draft Jan 2025 | Context of use + credibility activities |
| ICH M15 MIDD | ICH / EMA Step 5 | CHMP transmission Oct 2024+ | MAP/MAR, model risk for AI/ML models |
| AI regulatory strategy | MHRA | 30 Apr 2024 | Five principles + AIaMD reform path |
What does EMA expect when AI touches discovery or the dossier?
According to EMA’s scientific guideline page for AI in the medicinal product lifecycle, the reflection paper (EMA/CHMP/CVMP/83833/2023) was adopted by CHMP on 9 September 2024 and by CVMP on 11 September 2024 after a July–December 2023 public consultation.
EMA states the paper covers AI/ML at any step from drug discovery to post-authorisation. For discovery and nonclinical work, that includes AI used to replace, reduce, or refine animal models. For clinical stages, examples include patient selection, data recording, and analyses that later enter marketing-authorisation packages. Where AI/ML is expected to affect benefit–risk, EMA advises early regulatory support such as qualification of innovative methods or scientific advice.
The paper must be read with the EU AI Act, GDPR, cybersecurity rules, and medicines law. Sponsors should map which AI uses fall under EMA versus national competent authorities, because assessment intensity follows that remit.
How do the FDA–EMA good AI principles change R&D governance?
FDA’s Guiding Principles of Good AI Practice in Drug Development page (January 2026) confirms CDER and CBER developed 10 principles with EMA. They emphasize human-centric design, a risk-based approach, adherence to standards, clear context of use, multidisciplinary expertise, data governance and documentation, model design practices, risk-based performance assessment, life-cycle management, and clear essential information for users.
For EU/UK teams running global programmes, these principles are a shared vocabulary with FDA reviewers. A discovery model used only for internal target ranking still needs documentation if its outputs later influence clinical design or CMC claims. A model whose outputs support a primary analysis needs stronger credibility evidence.
What is FDA’s credibility framework for AI submissions?
In January 2025, FDA issued draft guidance titled Considerations for the Use of Artificial Intelligence to Support Regulatory Decision-Making for Drug and Biological Products. The agency frames a risk-based credibility assessment tied to a defined context of use (COU)—how the model answers a specific question of interest.
FDA’s accompanying announcement said the draft drew on experience with more than 500 AI-containing submissions since 2016 and more than 800 comments on earlier discussion papers. EU sponsors filing in the US should not treat COU as optional boilerplate: define the question, model influence, and decision consequence before locking protocols.
Where does ICH M15 fit for AI/ML models?
EMA’s ICH M15 Step 5 PDF on general principles for model-informed drug development explicitly lists artificial intelligence/machine learning among modelling and simulation methods used to generate MIDD evidence. M15 focuses on assessing that evidence—model risk, technical criteria, verification/validation/applicability, and Model Analysis Plans (MAPs) and Model Analysis Reports (MARs).
For discovery-to-clinic handoff, that means AI-derived predictions that later justify dose, enrichment, or sample-size claims should be planned and reported like other MIDD work, with overfitting and data-dependency risks called out as method-specific issues.
What must UK sponsors do under MHRA’s AI strategy?
On 30 April 2024, MHRA’s AI regulatory strategy press release set out how the agency applies the UK government’s five AI White Paper principles covering safe and secure systems, transparency and explainability, fairness, accountability and governance, and contestability and redress.
MHRA looks at AI as a regulator of products, as a public-service decision-maker, and as a consumer of third-party evidence. Where AI is used for a medical purpose, UK MDR 2002 typically applies. The agency also co-authored 10 Good Machine Learning Practice principles with FDA and Health Canada, covering representative data, independent train/test sets, human–AI team performance, and post-deployment monitoring.
UK sponsors using discovery AI that never becomes a medical device still face MHRA scrutiny when that AI shapes the evidence package for a medicine. Document training data provenance, performance metrics, and human oversight the same way you would for any critical analysis method.
What should EU/UK R&D teams change now?
Build an inventory of AI/ML tools from target ID through pharmacovigilance. For each tool, record COU, patient/regulatory impact, data sources, validation status, and the owner accountable for updates. Prefer early EMA scientific advice or MHRA innovation pathways when AI influences benefit–risk or label language.
Align global teams on the FDA–EMA 10 principles so the same model dossier can travel. Tie AI/ML analyses that support MIDD questions to ICH M15 MAP/MAR discipline. Do not assume a discovery-stage model is “out of scope” if its outputs later enter the CTD.
Related EU regulatory reading on NovaPharmaNews includes our coverage of PRAC June 2026 highlights, the 2026 EU drugs agency report, and Vanda’s EMA orphan positive opinion.
What remains unsettled for sponsors?
EMA’s reflection paper is current thinking, not a full technical guideline; formal AI guidelines may follow. FDA’s January 2025 AI credibility guidance was issued as draft. MHRA continues device-rule reform for AIaMD, including predetermined change control plans. Sponsors should treat these texts as the operating bar for 2026 dossiers while watching for finalized guidance and AI Act implementing detail.
Frequently Asked Questions
What does EMA say about AI in the medicinal product lifecycle?
EMA’s reflection paper on AI in the medicinal product lifecycle, adopted by CHMP on 9 September 2024, covers AI/ML from drug discovery through post-authorisation. It calls for a risk-based approach, bias control, and early scientific advice when AI affects benefit–risk.
How do FDA and EMA align on good AI practice in drug development?
In January 2026, FDA published 10 Guiding Principles of Good AI Practice in Drug Development developed with EMA. They stress human-centric design, clear context of use, data governance, risk-based performance assessment, and life-cycle management.
What should UK sponsors do under MHRA AI principles?
MHRA’s April 2024 AI strategy adopts five White Paper principles on safe systems, transparency, fairness, accountability, and contestability. Sponsors using AI as a medical device must also follow UK MDR 2002 and joint FDA–Health Canada–MHRA good machine learning practice principles.
Primary Sources
- EMA: Use of Artificial Intelligence (AI) in the medicinal product lifecycle
- FDA: Guiding Principles of Good AI Practice in Drug Development
- FDA: Considerations for the Use of Artificial Intelligence (draft guidance)
- EMA: ICH M15 Guideline on MIDD (Step 5 PDF)
- GOV.UK: MHRA AI regulatory strategy (30 April 2024)
- GOV.UK: Good machine learning practice guiding principles
This article follows our editorial standards. Report a correction via editorial contact.
