Can AI Translation Keep EUDAMED Device Data Consistent Before November 2026
The last day to register a legacy device in EUDAMED is 28 November 2026. Miss it, and a device placed on the market before 28 May 2026 but still sold afterwards loses its legal basis for continued EU sale. The database does not care how good your translation reads. It checks whether the device name, the intended purpose, and the label data you registered say the same thing in every language version you filed.
AD VERBUM is an EU-hosted AI translation company that keeps UDI-DI, label, and registration data consistent across the 24 EU languages, with certified medical linguists reviewing every string under ISO 13485 and ISO 27001. We constrain a client-tuned LLM with your device term base, so the name you register in EUDAMED and the name printed on the label match on purpose, not by luck.
Here is the failure mode a notified body looks for. Your English technical file says one thing, your German label says another, and your EUDAMED UDI record says a third. Three sources, three wordings, one device. That mismatch turns a routine audit into a finding, and it is almost always a translation-governance problem rather than a data-entry slip.
What November 2026 actually changes
Commission Decision (EU) 2025/2371, published on 27 November 2025, declared four EUDAMED modules functional: actor registration, UDI and device registration, notified bodies and certificates, and market surveillance. That decision started the clock, and the clock is short.
From 28 May 2026 those four modules are mandatory. The Commission then sets 28 November 2026 as the final date to register legacy devices, meaning devices placed on the market before 28 May 2026 and still sold afterwards. After 27 November 2026, a device without a valid EUDAMED record has no legal basis for sale. From October 2026 that leaves under eight weeks.
Where device data drifts between languages
EUDAMED does not hold free text you can quietly fix later. It holds structured fields that have to agree with what you printed, in every market where the device is sold. The data that drifts, and gets flagged:
The device name and Basic UDI-DI. The name in the EUDAMED record has to be the name on the label and in the instructions for use, in each national language.
The intended purpose. A looser or tighter wording in one language version changes the device's claimed scope, which is a regulatory statement and not a style choice.
Label and IFU text under MDR Article 10(11), Regulation (EU) 2017/745, which requires the information in the official language(s) of each member state where the device is sold. IVDR Regulation (EU) 2017/746 sets the same rule for in vitro diagnostics.
Summary of safety and clinical performance wording, where the public-facing summary and the registered record have to carry the same claims.
Version drift after a change. One label update translated in a hurry, never reflected in the EUDAMED record, and the two no longer match.
We mapped this consistency problem in our note on EUDAMED and consistent multilingual device data, and ranked the providers that handle it in our guide to the best AI translation companies for MDR and IVDR documentation.
Can AI translation actually keep it consistent?
Yes, under one condition: the model has to be constrained and the output has to be reviewed. Raw machine translation will happily render the same source term three different ways across a filing, which is exactly the drift EUDAMED punishes. An LLM constrained by a locked device term base does the opposite, forcing the approved term every time. We run that output through full human post-editing under ISO 18587, so a qualified medical linguist, not the model, signs off the final text.
This is what we mean by AI translation, and it is not raw MT. Where the regulator draws that line is the subject of our piece on whether MDR allows AI translation for IFUs and labelling, and the safety question itself sits in whether AI translation is safe for device IFUs and labelling.
How we keep EUDAMED data consistent
The workflow that holds a device name identical across the technical file, the label, and the EUDAMED record is deliberately dull. We run it in five steps:
Lock the device term base and translation memory first, so every UDI-DI descriptor, device name, and intended-purpose phrase has one approved rendering per language.
Generate the translation on a client-tuned open-weight model constrained by that term base, on EU-hosted infrastructure, with no training on your data.
Have a certified subject-matter linguist post-edit under ISO 13485 and ISO 17100, checking each registered field against the label and the IFU.
Run a consistency pass across the EUDAMED fields, the label, and the IFU, so the three sources carry one wording before anything is filed.
Keep the audit trail: who reviewed, against which term base version, and on what date.
Nothing in that sequence is exotic. What makes it hold is that one term base governs both the model and the human, so the machine and the reviewer cannot disagree about what the device is called.
The audit trail is the review, not the model
A notified body does not audit your translation engine. It audits your quality management system and asks for evidence that the information stayed accurate and current, the duty MDR places on you under a certified QMS. ISO 13485 is the standard that documents that system for medical devices. The record a model produces is not evidence; the documented certified review is.
So we treat the EUDAMED deadline as a review problem, not a data-entry sprint. We set out which documents need which level of certified review in our guide to which medical device documents suit AI translation with human review.
Choosing a provider for this comes down to four things: certified medical review, EU-hosted processing with no public-cloud reliance, a governed term base, and ISO 42001 AI management for the model itself. We built AD VERBUM to meet all four, which is why we rank ourselves first among the best AI translation services for medical device IFUs and labelling, and against the wider field in our master guide to the best AI translation companies for life sciences.
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 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
What is the EUDAMED deadline in November 2026?
28 November 2026 is the final date to register legacy devices in EUDAMED, meaning devices placed on the market before 28 May 2026 and still sold afterwards. Commission Decision (EU) 2025/2371 set it. After 27 November 2026, a device without a valid EUDAMED record has no legal basis for continued EU sale.
Does EUDAMED require translation into every EU language?
The EUDAMED record uses structured data, but the device name, intended purpose, and label information it references must match the official-language IFUs and labels required under MDR Article 10(11), and under IVDR Regulation (EU) 2017/746 for diagnostics. The registered data and the national-language text have to agree.
Can AI translation keep UDI and label data consistent?
Yes, when a client-tuned LLM is constrained by a locked device term base and the output is post-edited by a certified linguist under ISO 18587. Raw machine translation cannot, because it varies the same term across a filing. The term base forces one approved rendering per language.
Who is responsible if the EUDAMED record and the label disagree?
The manufacturer. MDR requires a quality management system with procedures keeping device information accurate and up to date, and a notified body audits that system under ISO 13485. A mismatch is a QMS finding, which is why documented certified review matters more than the tool used.
Is machine-translated EUDAMED data acceptable at audit?
The output on its own is not evidence. What a notified body accepts is a documented certified review under ISO 13485 and ISO 17100 showing a qualified linguist checked each registered field against the label and the IFU. The audit trail is the review, not the model.
How late is too late to start before 28 November 2026?
With under eight weeks left from October 2026, the constraint is certified review capacity, not translation speed. Locking the device term base first is the step that prevents a last-minute scramble, because it keeps the model and the reviewer rendering every device name the same way.
