Does MDR Allow AI Translation for IFUs and Labelling
MDR names no translation tool, and it bans none. Regulation (EU) 2017/745 asks something narrower: can you prove the translated instructions for use and label are accurate, current, and produced under a controlled process? AI translation clears that bar when a certified linguist post-edits the output and the files stay inside your quality management system.
AD VERBUM is an EU-hosted AI translation company that runs every medical device IFU and label through certified medical linguists, using a client-tuned LLM constrained by your Translation Memory and Term Base, with no public-cloud processing and no training on client data. That's the shape MDR accepts. The model drafts, a qualified human signs off, and the evidence sits in your technical file for the notified body to read.
What MDR actually requires
MDR sets two obligations that decide this question, and they work together.
Article 10(11) requires the information supplied with a device, the IFU and the label under Annex I Section 23, in the official Union language(s) of each member state where the device is made available. Every country sets its own languages. Sell across the whole EU and you can face all 24.
Article 10(9) requires the manufacturer to operate a quality management system that keeps regulatory documentation, including translated IFUs and labels, accurate and current across the device lifecycle. Read the two together and the rule is plain: the language must be right, and you must be able to show how you got it right.
The IVDR (Regulation (EU) 2017/746) carries the same duty for in vitro diagnostic devices. Neither text mentions machine translation, neural MT, or large language models. They regulate the outcome and the process, not the tool.
Where AI translation fits the QMS duty
A quality management system doesn't care which software produced a first draft. It cares whether a competent person is accountable for the final text, whether the process is documented, and whether you can reproduce the result. Those are exactly the controls a certified post-editing workflow supplies.
ISO 18587 defines full human post-editing of machine and AI output, and it puts a qualified post-editor on the hook for the final translation. ISO 13485 sets the device QMS and supplier controls. ISO 17100 adds the qualified-linguist and independent-revision step. Stack them and AI-drafted output becomes a controlled process with a named human owner, which is what Article 10(9) asks for.
The compliance gap is unreviewed output, not AI
The risk MDR guards against is a wrong instruction in a document a patient or clinician relies on. A mistranslated contraindication, a dosage figure dropped in the target language, a warning softened by a literal rendering. Those failures happen when raw machine output ships without a qualified review, whatever engine produced it.
A notified body auditing your technical file looks for the review record, not the model name. If a certified linguist post-edited the IFU, signed off, and the correction history is in the file, the tool that drafted it is your business. If nobody reviewed it, you have a gap, and that gap exists whether the draft came from a junior translator or an LLM.
So the honest answer to whether MDR allows AI translation is that MDR never asked about AI. It asked whether your process produces accurate, current, controlled language. Skip the human review and you fail that test. Keep it, and the model is just a faster first draft. The same logic applies to whether AI translation is safe for medical device IFUs.
What AI translation means for a device file
Not all AI translation is equal, and the distinction matters here. Raw neural MT translates segment by segment with no view of the whole document, so it loses cross-references and terminology consistency across a 60-page IFU. That's not what we mean by AI translation.
AI translation for regulated content, in the sense we use, is a large language model reading the document in context, constrained by your approved terminology, then corrected by a certified subject-matter linguist. We call the combination AI+HUMAN hybrid translation, and it's the core of what AI translation actually means for regulated work. The LLM handles volume and consistency, the human owns accuracy and compliance. On a device label, that difference is the space between a first draft and a submission.
EU data residency and ISO 27001
IFUs and technical files carry design detail you don't want on a public cloud API. Under GDPR and MDR confidentiality expectations, where the text is processed is part of the compliance question.
We at AD VERBUM run AI translation on self-hosted, open-weight models on EU infrastructure, certified to ISO 27001 for information security and ISO 42001 for AI management. Client content is not sent to third-party model APIs and is never used to train models. The same EU-hosted control covers our AI translation services for device IFUs and labelling and our wider AI translation for life sciences. For a device file heading to a notified body, that keeps the data trail as clean as the translation record.
A compliant AI translation workflow for device IFUs
Here's the sequence that keeps AI-drafted IFUs and labels inside MDR's QMS requirement:
Lock the device Term Base and Translation Memory so the model generates against approved terminology, not guesses.
Run constrained LLM generation on EU-hosted infrastructure, with no client data leaving your controlled environment.
Apply full human post-editing by a certified medical linguist under ISO 18587 and ISO 13485.
Check IFU, label, and EUDAMED data for consistency across every target language before release.
Keep the correction history, reviewer sign-off, and version record in the technical file for the notified body.
Follow that and AI translation isn't a compliance risk. It's a documented process that produces the same audit trail MDR expected from human-only work, faster. The approach carries across regulated verticals, from device labelling to EMA eCTD submissions and pharma clinical trials.
Our medical device translation services
Our translation services for regulated sectors run on ISO 27001 and ISO 42001 certified, EU-hosted infrastructure, with no reliance on public cloud tooling for core processing. Every project runs through our AI+HUMAN hybrid workflow: we ingest client Translation Memories and Term Bases first, our proprietary LLM-based LangOps System generates output constrained by client terminology on client-tuned open-weight models, and our certified subject-matter experts review for technical accuracy and regulatory compliance. Our QA is aligned to ISO 17100 and ISO 18587, with sector-specific requirements such as ISO 13485 medical-device QMS controls and MDR Article 10(11) language scope applied where relevant. We serve Life Sciences, Legal, Finance, Defense, and Manufacturing clients across 150+ languages with 3,500+ subject-matter linguists. For teams managing audit-sensitive content, contact us to discuss your security and compliance requirements directly.
FAQ
Does MDR require certified translation for IFUs and labels?
MDR Article 10(11) requires IFUs and labels in the official language(s) of each member state where the device is sold, and Article 10(9) requires a QMS that keeps them accurate and current. It doesn't mandate a specific certification, but ISO 13485 and ISO 17100 give the documented, qualified process a notified body expects to see.
Is AI translation allowed under MDR?
Yes. MDR names no translation technology and prohibits none. It requires accurate, current language produced under a controlled process, so AI translation is allowed when a certified linguist post-edits the output under ISO 18587 and the review is documented in the technical file.
What is the difference between AI translation and machine translation for devices?
Raw machine translation renders text segment by segment with no document context. AI translation as we deliver it uses a large language model reading the full document, constrained by your Term Base, then corrected by a certified medical linguist. For a multi-page IFU, that difference shows up as terminology and cross-reference consistency.
Who is responsible if an AI-translated IFU has an error?
The manufacturer holds regulatory responsibility under MDR, and within the translation process the certified post-editor named under ISO 18587 is accountable for the final text. That named human sign-off is why unreviewed machine output fails an MDR QMS and post-edited output does not.
Does AI translation keep IFU data secure?
It can, if the processing stays off public cloud APIs. We run AI translation on self-hosted models on EU infrastructure certified to ISO 27001, with no client data used to train models, so IFU and technical-file content stays inside a controlled, GDPR-aligned environment.
Does the same rule apply to IVDR devices?
Yes. IVDR (Regulation (EU) 2017/746) carries the same language and QMS obligations for in vitro diagnostic devices. AI translation with certified human review meets the IVDR IFU and labelling requirement the same way it meets MDR's.