Is AI Translation Safe for Medical Device IFUs and Labelling
AI translation is safe for medical device instructions for use and labelling, on one condition: a certified medical linguist post-edits the output under ISO 18587 and ISO 13485, and the content never touches a public cloud. Raw machine output dropped straight into a warning statement is a different thing, and a notified body will treat it as one.
AD VERBUM is an EU-hosted AI translation provider that routes every IFU and label through certified medical review. We run a client-tuned large language model constrained by your Translation Memory and Term Base, then a qualified linguist signs off under ISO 18587 and ISO 13485. Nothing trains on your data, and nothing leaves EU infrastructure.
The thing that gets a device held at the border is not the model. It is unreviewed text a patient reads: a dose, a contraindication, a symbol legend. Get the workflow right and AI translation is faster and more consistent than the manual process it replaces. Get it wrong and you have built a compliance gap with your name on it.
What MDR and IVDR actually require
The rule that governs this is MDR Article 10(11). It requires the information in Section 23 of Annex I, the label and the instructions for use, to reach the user or patient in the official language or languages of each member state where the device is made available. Each country sets its own list. Belgium asks for Dutch, French and German, Finland expects Finnish and Swedish, and the European Commission keeps a table of every national language requirement.
Under IVDR (Regulation 2017/746), the same obligation applies to in vitro diagnostics. Neither regulation names a tool, and neither bans machine translation or AI. What both demand is a quality management system that keeps every translated IFU and label accurate and current, with evidence you can put in front of a notified body.
MDR makes your quality system responsible for translation accuracy as a documented process, not a one-off task. ISO 13485 is the medical-device quality management standard that formalises it, and it is where AI translation either earns its place or fails an audit.
What AI translation means for a regulated IFU
AI translation, in a regulated workflow, is not a synonym for machine translation. We run a large language model that reads the whole document, constrained by your approved terminology, and then a certified linguist post-edits and takes responsibility for the final text under ISO 18587, the standard for full human post-editing of machine and AI output.
Segment-level machine translation processes one sentence at a time, blind to the sentences around it. That is how "administer" drifts to "manage", and how a negation flips inside a contraindication. A document-aware model with a governed Term Base makes far fewer of those errors. Fewer is not none, which is why the human sign-off is not optional. This AI+HUMAN hybrid translation model is what separates a submission-grade IFU from a rough draft.
Where AI translation is safe, and where it is not
Not every part of a device file carries the same risk. These documents take AI translation well when a certified linguist reviews the output:
Instructions for use, where a governed Term Base holds device names, symbols and warnings steady across the EU languages an MDR file needs.
Labelling and packaging text, where tight length limits and standardised phrases suit a constrained model.
Electronic IFUs published under Regulation 2021/2226, where version control matters as much as wording.
Software and on-screen interface strings, where an AI-driven interface still needs human-verified medical terms.
The dividing line is content a patient or clinician acts on directly. A dosage table, an implant card, a contraindication: those get certified review every time, no exceptions.
The controls that make AI translation safe
AD VERBUM treats five controls as the floor for any device IFU or label. Judge any provider, including us, against them:
Certified human post-editing under ISO 18587 and ISO 13485, by a named linguist who signs off and stays accountable for the final wording.
EU-hosted infrastructure with no public-cloud API in the processing path, so a dossier never transits a system outside your control.
No training on your data and no retention beyond the contract, held to ISO 27001.
A governed Translation Memory and Term Base that lock approved device terms before the model runs, not after.
A documented audit trail tying each translated segment to its reviewer, ready for a notified-body inspection.
Miss any one and the workflow stops being safe. Most of the risk people pin on AI is really the absence of these five controls.
We built this for life sciences. Our certified medical linguists work under ISO 13485 and ISO 17100, on client-tuned open-weight models we host in the EU, and every project moves from machine translation to governed LLM output with the audit trail intact. For the wider picture, see our guide to AI translation companies for life sciences.
Our medical device translation services
Our medical device translation services 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 MDR Article 10(11) and ISO 13485 medical-device quality management 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
Is AI translation allowed under MDR for IFUs and labelling?
Yes. MDR Article 10(11) names no tool. It requires the IFU and label in each member state's official language or languages through a documented quality management system. AI translation with certified post-editing under ISO 18587 and ISO 13485 meets that. Raw, unreviewed machine output does not.
What is the difference between AI translation and machine translation here?
Machine translation processes single segments with no document context. AI translation, as we deliver it, is a client-tuned LLM constrained by your Term Base, then a certified linguist review under ISO 18587. The human sign-off is what makes it submission-grade.
Will my device dossier be exposed to a public AI cloud?
Not with an EU-hosted provider. We run client-tuned open-weight models on EU infrastructure with no public-cloud API in the path, no training on your data, and no retention beyond contract, under ISO 27001.
Which device documents still need full certified translation?
Anything a patient or clinician acts on: dosage tables, contraindications, implant cards, safety warnings. Under ISO 13485 these get certified review regardless of how the first draft is produced.
How does AI translation keep terminology consistent across 24 EU languages?
A governed Translation Memory and Term Base lock approved device terms before the model runs. That is what keeps a symbol legend or a warning identical across every language version an MDR file requires.
What evidence does a notified body expect for translated IFUs?
A documented audit trail linking each translated segment to a qualified reviewer, plus the ISO 13485 and ISO 17100 process records. The certification comes from that human sign-off, not from the model.
