Artificial intelligence is increasingly being introduced not only by pharmaceutical companies, but also by regulatory authorities. A new IFPMA position paper, published on 9 November 2026, examines what responsible use should look like when AI supports regulatory review, scientific assessment, inspections, document preparation and other authority activities.
The paper’s central message is important for pharmaceutical companies: AI can improve regulatory capacity and efficiency, but human regulatory judgement, protection of sponsor data, source verification and accountability must remain central.
The Paper Focuses on Regulators’ Use of AI
The position paper was published by the International Federation of Pharmaceutical Manufacturers and Associations, or IFPMA.
Its scope is specific.
It examines how national regulatory authorities may use AI when conducting regulatory activities related to medicinal products.
Examples include:
- internal regulatory operations;
- document and information retrieval;
- regulatory submission review;
- scientific assessment;
- inspection support;
- preparation of regulatory outputs;
- analysis of complex regulatory data.
The paper does not primarily address AI embedded in medicinal products or AI systems operated by pharmaceutical manufacturers.
Those uses involve separate regulatory considerations.
Regulators Are Already Moving from Strategy to Implementation
IFPMA notes that regulatory authorities are at different stages of AI adoption.
Some have published strategies or action plans.
Others are testing or deploying AI tools in operational workflows.
The paper refers to examples involving:
- PMDA use of AI for document retrieval, summarisation and translation;
- EMA and HMA coordination through the Network Data Steering Group;
- MHRA regulatory review and GMP-compliance tools;
- FDA use of AI for information retrieval, regulatory data analysis and workflow support;
- ANVISA use of AI to support assessment of impurity-qualification information.
These examples show that AI is beginning to influence the environment in which pharmaceutical submissions, inspections and regulatory interactions take place.
Human Regulatory Judgement Must Remain Central
IFPMA states that AI may support:
- data analysis;
- evidence review;
- preparation of regulatory outputs;
- identification of information requiring further attention.
However, final regulatory responsibility must remain with the competent authority and its qualified assessors and inspectors.
This creates an important boundary:
AI may support the regulatory process, but it should not become the unaccountable decision-maker.
For pharmaceutical companies, this matters because an AI-supported regulatory question or finding can still have significant consequences.
The responsible human assessor should therefore remain capable of:
- understanding the AI-supported output;
- reviewing its evidentiary basis;
- challenging it;
- confirming whether it is appropriate;
- taking responsibility for the resulting regulatory action.
AI-Generated Regulatory Findings Should Be Verified Against Source Documents
One of the strongest recommendations concerns source verification.
Where AI contributes to:
- regulatory findings;
- requests for information;
- scientific assessments;
- inspection questions;
- regulatory conclusions,
the resulting output should be verified against the relevant authoritative documentation.
This is important because an unsupported AI-generated question can create:
- unnecessary sponsor workload;
- avoidable delays;
- incorrect regulatory conclusions;
- loss of confidence in the review process.
A defensible regulatory workflow should therefore look like:
AI-Supported Analysis → Source Verification → Qualified Regulatory Judgement → Regulatory Output
Assurance Should Be Proportionate to the Intended Regulatory Use
The paper supports a fit-for-purpose and risk-based approach.
The required level of:
- oversight;
- evaluation;
- validation;
- documentation
should depend on:
- the intended use of the AI;
- the regulatory context;
- the degree to which AI influences the activity;
- the possible impact on patients, public health and regulatory outcomes;
- the complexity of the AI system where relevant.
This avoids two inappropriate extremes:
- treating every AI use as equally critical;
- treating AI as a low-risk productivity tool regardless of its regulatory influence.
The relevant question is not only:
Is AI being used?
It is:
What regulatory decision can the AI influence, and what could happen if its output is wrong?
Sponsor Data Should Not Be Used to Train Regulatory AI Without Authorisation
The paper gives particular attention to confidential and proprietary information.
Regulatory authorities routinely receive:
- clinical data;
- manufacturing information;
- CMC information;
- commercially confidential information;
- proprietary scientific evidence.
IFPMA states that sponsor or other proprietary data should not be used for:
- training;
- fine-tuning;
- improving AI systems;
- purposes outside the original regulatory use
without appropriate authorisation or agreement and suitable safeguards.
This is one of the paper’s most important practical positions.
Data from Different Sponsors Should Remain Segregated
Where a regulatory AI system processes information belonging to multiple sponsors or data owners, the paper recommends appropriate data segregation.
Information belonging to one company should not unintentionally influence outputs concerning another company.
This creates important technical and governance questions:
- Can one sponsor’s data enter the context used for another sponsor’s assessment?
- Are model prompts and outputs isolated?
- Are retrieval indexes segregated?
- Can proprietary information affect future model behaviour?
- Can cross-sponsor leakage be detected?
For pharmaceutical companies, assurances about data separation may become an important part of trust in regulator-operated AI systems.
Transparency Is Needed When AI Influences Regulatory Assessment
IFPMA calls for clarity concerning:
- when AI is used;
- how it is used;
- the purpose of the AI-supported activity;
- the role AI plays in regulatory assessment and decision-making.
This does not necessarily mean disclosing every technical detail of a regulatory authority’s internal system.
However, stakeholders should have sufficient visibility to understand when AI has materially contributed to an assessment or regulatory output.
Transparency becomes especially important where AI-generated analysis contributes to:
- questions raised to an applicant;
- inspection findings;
- scientific conclusions;
- regulatory requests.
The AI Tool, Model Version and Assessment Record Should Be Traceable
Where AI contributes to an assessment, IFPMA recommends appropriate traceability of:
- the AI tool;
- the model version;
- the relevant assessment record;
- the conditions under which the output was generated.
This could support:
- reproducibility;
- auditability;
- verification;
- investigation of inconsistent outputs;
- reconstruction of regulatory reasoning.
This is particularly important because two regulators using different model versions, configurations or prompts could obtain different outputs from similar information.
Regulatory AI Outputs Should Remain Traceable to the Sponsor Dossier
Where AI extracts, reorganises or summarises information from a regulatory submission, the resulting output should remain connected to the original dossier source.
A practical architecture could therefore be:
Sponsor Dossier
↓
AI Retrieval or Extraction
↓
Source-Linked Regulatory Output
↓
Human Verification
↓
Assessment Record
This is substantially more defensible than allowing an AI system to produce unsupported regulatory narrative without evidence links.
Data Quality and Provenance Remain Fundamental
The reliability of regulatory AI depends on the data used for:
- training;
- validation;
- operation.
IFPMA highlights:
- data quality;
- relevance;
- representativeness;
- provenance;
- integrity.
Poor-quality or poorly governed data can result in inconsistent or misleading regulatory outputs even when the AI model itself is technically sophisticated.
The paper therefore reinforces a familiar principle:
Regulatory AI credibility begins with data credibility.
Lifecycle Management Is Necessary
AI systems can change because of:
- model updates;
- training-data changes;
- configuration changes;
- new usage patterns;
- changes in regulatory data;
- supplier updates.
IFPMA therefore identifies the need for:
- version management;
- record keeping;
- ongoing monitoring;
- periodic re-evaluation;
- lifecycle management.
The objective is to ensure that the system continues to remain fit for its regulatory purpose over time.
International Convergence Is Becoming Essential
Regulatory authorities are developing AI capabilities at different speeds and under different legal frameworks.
Without alignment, pharmaceutical companies could face:
- different terminology;
- different expectations for AI-supported submissions;
- inconsistent transparency;
- different approaches to sponsor-data use;
- different validation and oversight expectations.
IFPMA therefore supports collaboration on:
- common terminology;
- use-case classification;
- evaluation approaches;
- documentation practices;
- performance expectations;
- capacity building.
The paper points to FDA–EMA collaboration and ICMRA discussions as useful foundations for greater convergence.
What Pharmaceutical Companies Should Watch
As regulators increase their use of AI, companies should monitor:
- whether authorities disclose AI-supported review activities;
- how sponsor data are protected;
- whether AI outputs remain traceable to submitted evidence;
- how inconsistent or incorrect AI-supported findings can be challenged;
- whether regulatory authorities use common terminology and assurance approaches;
- how AI affects inspection selection, dossier review and requests for information.
Pharmaceutical organisations may also need more structured, machine-readable and source-traceable submissions as regulators increase AI-supported review.
Why This Position Paper Matters
Much of the regulatory discussion has focused on how pharmaceutical companies should use AI.
This paper addresses the other side of the regulatory interface:
How should regulators use AI when reviewing companies, submissions and evidence?
The central principle is balanced:
Regulators can benefit from AI, but the same expectations concerning evidence, traceability, confidentiality, verification and accountability remain essential.
Regulatory Status Note
The document is an IFPMA industry position paper published on 9 November 2026.
It is not guidance issued by FDA, EMA, PMDA, MHRA or another regulatory authority and does not create binding requirements.
It presents the innovative pharmaceutical industry’s position and recommendations regarding the use of AI by regulatory authorities.
Sources
IFPMA – The use of artificial intelligence in regulatory activities: key areas and considerations, published 9 November 2026:
https://www.ifpma.or … -and-considerations/
Direct link to the IFPMA position paper:
https://www.ifpma.or … iderations_IFPMA.pdf
You can see also:
IFPMA: Pharmaceutical Manufacturing Can Be a Controlled Environment for Responsible AI Adoption:
https://www.ifpma.or … tion-and-regulation/