AI in Regulatory Translation: What EMA, FDA, and PMDA Say About Machine Translation

Regulators around the world are watching closely as the translation industry experiments with new tools. For companies submitting dossiers to authorities such as the European Medicines Agency (EMA), the U.S. Food and Drug Administration (FDA) and the Pharmaceuticals and Medical Devices Agency (PMDA, Japan), the use of machine translation (MT) or AI-assisted translation can seem like an appealing shortcut — especially under tight deadlines or for multilingual submissions. But what do these agencies actually say about machine translation, and under what conditions (if any) is it acceptable?

In a regulatory context, translation is not just a convenience, but a critical component of safety, clarity and compliance. Submission materials — from clinical study reports to product labeling and patient-facing documents — must meet stringent requirements for accuracy, consistency, and traceability across different languages. Any mistranslation, ambiguous phrasing or formatting error can lead to serious consequences: regulatory queries, delayed approval, or even outright rejection.

In principle, regulators focus less on how a translation was produced and more on the quality and compliance of the final text. For the EMA, while there is no explicit public regulation that bans the use of AI-assisted translation, the agency demands that all translations submitted under the centralized procedure adhere to predefined templates (such as the QRD template for product information) and undergo rigorous linguistic review. Every language version must incorporate any Member State comments before finalization.

Similarly, in pharmacovigilance — for example Individual Case Safety Reports (ICSRs) — the EMA’s Good Pharmacovigilance Practices (GVP) require either the original verbatim text or an accurate translation. Some recent reflections around AI use in the medicinal product lifecycle emphasize that if AI is used, human expert review and auditability remain essential.

In the United States, the FDA has been even more explicit about its stance. According to industry-wise guidance, raw machine translation alone is not acceptable for regulatory or clinical trial materials such as informed consent forms (ICFs) or safety documentation. Translations must be produced (or at least thoroughly reviewed) by qualified human linguists, with complete documentation: translator credentials, revision history, quality-assurance checks, and so on.

For Japan and the PMDA, the demand is for full translations into Japanese for all core regulatory documents (e.g. CTD modules, Clinical Study Reports, Investigator Brochures, safety updates). Reviewers expect not only linguistic accuracy but also conformity with Japanese regulatory terminology, scientific style, and cultural context.

Given these realities, how can AI or MT tools be incorporated — if at all — in a compliant, risk-managed translation workflow? Many translation providers and localization-tech consultancies argue that AI is acceptable when used responsibly: as a first draft, followed by human post-editing, robust QA, terminology management, and full traceability. Standards such as ISO 18587 (post-editing of machine translation) or ISO 13485 (quality systems for medical device translation) provide frameworks to ensure that output quality meets regulatory expectations.

Therefore, AI in regulatory translation should not be framed as a replacement for human expertise, but as a support tool — one that, when embedded in a certified, quality-controlled process, can help manage large multilingual workloads without sacrificing compliance or patient safety.

Still, even with the best workflows, AI-assisted translation may not always be appropriate. Sensitive materials like adverse-event reports, patient safety communications, safety labeling, and documents intended for regulatory review — especially first submissions of new products — often demand the highest level of human oversight. For those, many translation experts recommend sticking with human-led translation and review to minimize risk.

The evolving regulatory landscape suggests a cautious but open-minded approach. The EMA’s recent “AI workplan” acknowledges the growing role of digital tools in medicines regulation. What remains constant is the expectation that translation — regardless of the tools used — results in accurate, clear, compliant, and traceable documents.

In practice this means that if your company chooses to use AI or MT in translation workflows, you must ensure: an experienced regulatory-savvy human team reviews and post-edits all output; that terminology is consistent with official standards (e.g. using correct terms from controlled vocabularies such as MedDRA); that audit trails are maintained; and that final versions conform to the relevant template and formatting requirements (QRD for EMA, Japanese-reg style for PMDA, etc.).

In short: AI and MT are not banned — they are tools. What regulatory authorities want is a final product that meets their exacting standards. At the intersection of speed, cost-efficiency, and compliance, the responsible use of technology — guided by human expertise — can deliver both.