The European medicines regulatory network is putting data quality and governance at the centre of its artificial intelligence strategy.
On 12–13 November 2026, the Heads of Medicines Agencies (HMA) and European Medicines Agency (EMA) will hold a new multistakeholder AI and Data Forum titled:
Build trustworthy AI on strong data foundations.
The event (with the public live stream) will bring together regulators and stakeholders to discuss the use of artificial intelligence and data across the medicines lifecycle.
The title itself sends an important message.
The regulatory discussion around pharmaceutical AI is increasingly moving beyond the question:
“Is the AI model sufficiently accurate?”
towards a broader question:
“Are the data, governance and infrastructure surrounding the AI sufficiently trustworthy to support regulated use?”
AI and Data Are Being Treated as One Regulatory Problem
EMA describes data and AI as increasingly interdependent.
High-quality data provide the foundation for scalable and trustworthy AI, while AI can in turn help organisations obtain more value from increasingly large and complex regulatory and health datasets.
This creates an important relationship:
Data Quality → Data Governance → Interoperability → AI Performance → Regulatory Trust
For pharmaceutical companies, this means that AI governance cannot be separated from data governance.
A technically advanced model may still be unsuitable for a regulated use if the underlying data are:
- incomplete;
- poorly structured;
- inconsistently defined;
- not sufficiently traceable;
- not representative of the intended use;
- difficult to exchange between systems;
- of uncertain provenance.
In other words:
better AI cannot compensate automatically for weak data foundations.
What the HMA/EMA Forum Will Address
According to EMA, the forum has several objectives.
It will:
- showcase current AI and data applications across the medicines lifecycle;
- discuss the evolving European regulatory landscape for AI and data;
- examine data foundations, governance, quality and interoperability;
- strengthen collaboration between regulators and stakeholders;
- provide input to the Network Data Steering Group workplan and the European medicines network strategy to 2028.
This is broader than a conventional AI workshop.
The focus is not only on algorithms.
It includes the regulatory data infrastructure required to make AI reliable, scalable and useful.
The Forum Is Part of a Larger HMA/EMA AI Strategy
The event is coordinated by the HMA/EMA Network Data Steering Group (NDSG), together with the CHMP Methodology Working Party and the Accelerating Clinical Trials in the EU initiative.
The NDSG workplan for 2026–2028 is particularly important background.
The workplan defines how the European medicines regulatory network intends to expand the use of data and AI in medicines regulation.
Planned activities include:
- development of data analytics and data standardisation frameworks;
- further development of real-world evidence infrastructure;
- implementation of the European Health Data Space;
- expansion of access to regulatory data assets;
- development of AI guidance for clinical development and pharmacovigilance;
- regulatory experimentation and capability building.
AI is therefore becoming part of a broader European regulatory data strategy rather than remaining a standalone technology initiative.
Why Data Standardisation Matters for AI
Artificial intelligence performs best when information is consistently structured and interpretable.
Pharmaceutical organisations, however, often operate complex data environments involving:
- laboratory systems;
- manufacturing systems;
- quality-management systems;
- clinical databases;
- regulatory repositories;
- pharmacovigilance systems;
- master-data platforms;
- unstructured documents.
The same concept may be represented differently across sites, systems or countries.
For example, products, processes, deviations, equipment, substances or regulatory events may use different terminology and metadata structures.
For conventional reporting this already creates difficulties.
For AI it becomes even more important.
An AI model can analyse large volumes of information, but if the underlying information does not have consistent meaning, the model may connect data incorrectly or generate misleading conclusions.
This is why interoperability and standardisation are increasingly becoming AI topics.
Data Provenance May Become as Important as Model Performance
For regulated pharmaceutical AI, it may not be enough to know that a model generated a particular result.
The organisation may also need to understand:
- where the underlying data originated;
- which version of the data was used;
- how the data were transformed;
- whether the source was controlled;
- whether important information was excluded;
- whether the data remain representative of the intended use.
This can be described as data provenance.
For some AI applications, confidence in the output will depend directly on confidence in this data lineage.
A useful assurance chain may therefore become:
Source Data
↓
Controlled Transformation
↓
Verified Dataset
↓
AI Model
↓
AI Output
↓
Human / Regulatory Decision
Each layer can affect the credibility of the final result.
Trustworthy AI Is Not Only a Model Validation Question
Much of the pharmaceutical AI discussion has focused on validation of the model.
Important questions include:
- Is the model sufficiently accurate?
- How should independent test data be selected?
- How should robustness be demonstrated?
- How should drift be monitored?
- When should revalidation occur?
These questions remain important.
But the HMA/EMA forum highlights another dimension.
Even a well-performing model depends on a broader control environment.
A mature AI assurance framework therefore needs to consider:
Data → Model → System → User → Process → Decision
rather than validating the algorithm in isolation.
Data Governance Becomes Part of AI Governance
For pharmaceutical companies, data governance may increasingly become one of the main components of AI governance.
Relevant controls can include:
- data ownership;
- defined data sources;
- data-quality criteria;
- metadata standards;
- master-data governance;
- data lineage;
- access control;
- retention;
- security;
- change management;
- interoperability standards.
This is especially important when AI applications combine information from multiple systems.
For example, an AI system supporting Quality Management could potentially analyse:
- deviations;
- CAPAs;
- complaints;
- audit findings;
- laboratory results;
- manufacturing trends;
- supplier information.
The value of the AI may depend less on the sophistication of the language model and more on whether these sources can be connected reliably.
AI Regulatory Strategy Is Moving Toward Infrastructure
This forum also illustrates an important evolution in regulatory thinking.
Earlier AI discussions frequently concentrated on broad principles such as:
- transparency;
- fairness;
- explainability;
- human oversight;
- accountability.
Those principles remain important.
However, regulators are increasingly moving toward practical implementation questions:
- Which data should AI use?
- How should those data be governed?
- How should data quality be demonstrated?
- How can different systems exchange information reliably?
- How can AI operate consistently across a regulatory network?
This is a transition from:
AI principles
towards:
AI infrastructure and regulatory implementation.
A Strong Message for Pharmaceutical Companies
For companies developing AI strategies, the practical implication is straightforward.
The AI programme should not start only with:
“Which AI model should we buy?”
It should also ask:
- Do we know which data the AI will use?
- Are those data sufficiently controlled?
- Are definitions consistent across systems?
- Can the source of every important data element be reconstructed?
- Are data-quality problems visible?
- Can information be exchanged between systems without losing context?
- Who owns the data?
This can lead to a more realistic implementation sequence:
Intended Use
↓
Data Requirements
↓
Data Quality and Governance
↓
AI Model and Architecture
↓
Validation / Credibility Assessment
↓
Controlled Deployment
↓
Lifecycle Monitoring
Relevance to GMP
The November forum is not specifically a GMP workshop and should not be interpreted as new GMP guidance.
However, its central message is highly relevant to GMP applications of AI.
Existing GMP expectations already place strong emphasis on:
- data integrity;
- traceability;
- controlled records;
- validated systems;
- defined responsibilities;
- change control.
AI does not reduce the importance of these principles.
It can increase it.
When an AI system derives conclusions from thousands or millions of data points, understanding whether those data are reliable becomes fundamental to understanding whether the AI output can be trusted.
For GMP AI, one useful principle may therefore be:
The validated state of an AI-enabled process depends not only on the model, but also on the continued state of control of its data.
What to Watch During the November Forum
Several areas will be particularly worth following:
- how EMA and HMA define trustworthy AI in practical terms;
- expectations for data quality supporting AI;
- approaches to data interoperability and standardisation;
- examples of AI use across the medicines lifecycle;
- the relationship between AI governance and the evolving EU regulatory framework;
- future NDSG workplan priorities;
- possible new AI guidance or regulatory initiatives.
The forum may therefore provide an important indication of the next phase of European pharmaceutical AI regulation.
The first phase focused heavily on AI principles.
The next phase may increasingly focus on:
data foundations → practical governance → implementation → regulatory evidence.
Participation
The HMA/EMA Multistakeholder AI and Data Forum will take place on:
12–13 November 2026.
It will be held at the European Medicines Agency in Amsterdam and online.
EMA states that online participation is available through Webex registration.
In-person participation is by invitation through official stakeholder groups.
The event will also be broadcast live.
Regulatory Status Note
The HMA/EMA AI and Data Forum is a multistakeholder regulatory workshop and does not itself establish new regulatory requirements.
Its discussions and presentations should not be interpreted as formal EMA guidance unless subsequently incorporated into official regulatory documents.
However, the forum is directly linked to the HMA/EMA Network Data Steering Group and its 2026–2028 workplan, making it an important indicator of the future direction of European medicines regulation in data and artificial intelligence.
Sources
European Medicines Agency – HMA / EMA multistakeholder AI and Data forum: Build trustworthy AI on strong data foundations:
https://www.ema.euro … ong-data-foundations
EMA – Network Data Steering Group:
https://www.ema.euro … -steering-group-ndsg
EMA – NDSG Workplan 2026–2028: Data and Artificial Intelligence in Medicines Regulation:
https://www.ema.euro … es-regulation_en.pdf
EMA – Artificial Intelligence: