What "AI Translation" Means for Regulated Content, and Why It Includes Human Review
AI translation for regulated content means a large language model constrained by your approved terminology, then a certified human reviewer who signs off on the result. It is not raw machine translation, and it is not a general chatbot left to guess. For an Instructions for Use, a Summary of Product Characteristics, or a clinical protocol, that definition decides your exposure, because a notified body or the EMA holds you to the reviewed text, not the model that drafted it.
AD VERBUM is an EU-hosted AI translation company whose output is generated by a client-tuned open-weight LLM, then verified by certified subject-matter linguists under ISO 17100 and ISO 18587. We run that workflow on EU infrastructure, with no public-cloud APIs and no training on your data. An LLM plus a named human who takes responsibility is what AI translation has to mean once the content is regulated.
What AI translation actually is
Three different things get sold as AI translation, and only one of them belongs near a regulated document. The gap between them is where compliance problems start.
Raw neural machine translation. Deterministic, segment by segment, fast and cheap, with no view of the whole document. Useful for gist, wrong for a warning label.
A general chatbot prompt. Fluent, but unconstrained by your terminology and hosted on public cloud infrastructure you do not control.
An LLM with certified human review. A model constrained by your Translation Memory and Term Base, generating document-aware output, then post-edited and signed off by a qualified linguist.
We never call raw machine translation AI translation. The certified human at the end is the difference, not a marketing label on the same engine.
Why regulated content forces human review
ISO 18587 sets the requirements for full post-editing of machine output, and it names a qualified human post-editor who takes responsibility for the final text. The standard applies to any engine, an LLM included, and its revision expected in October 2026 broadens the scope to post-editing of non-human translation output. ISO 17100 adds a second-linguist revision on top.
Regulators reason the same way. MDR Article 10(11) requires that device Instructions for Use and labels stay accurate and current under the manufacturer's quality management system, and EMA submissions are checked against the QRD template. The audit trail that satisfies them is the documented human sign-off. The model is not a party to your file.
The document-level problem machines miss
Segment-level machine translation cannot see the document it is translating. So a controlled term drifts to two renderings across a 90-page dossier, a warning sentence loses its force, or a dosage unit vanishes in one segment and no engine flags it. An LLM with full-document context handles far more of this, which is why AI translation outperforms classic machine translation on regulated content. It still does not reach submission accuracy on its own, which is the whole reason the certified reviewer stays in the workflow.
How AD VERBUM's AI translation works
Our AI+HUMAN hybrid translation runs as a fixed sequence, so every regulated project produces the same evidence:
We ingest your Translation Memory and Term Base first, so the model starts from your approved wording.
Our LangOps System generates output on client-tuned open-weight models, constrained by that terminology, on EU-hosted infrastructure.
A certified subject-matter linguist post-edits under ISO 18587 and takes responsibility for the result.
We run QA aligned to ISO 17100, then hand back a translation with a documented review trail and no client data left in a public cloud.
"AI translation vs human translation" is the wrong split
Buyers often frame the choice as AI translation versus human translation. For regulated content that is a false split. The real comparison is AI translation with certified human review against unreviewed output, whatever produced it. Human-only work without a documented second-linguist revision fails an audit for the same reason unreviewed machine output does. What clears a notified body or the EMA is the named reviewer and the record, not the tool that drafted the first version.
What to ask a provider who says "AI translation"
Before you accept the label, put four questions to any provider, including us:
Is every output post-edited by a certified linguist under ISO 18587, or is it raw model output?
Where is the model hosted, and is client data ever used to train it?
Is there a second-linguist revision under ISO 17100?
Is the AI workflow governed under ISO 42001, with the audit trail the EU AI Act expects?
The EU AI Act frames the same concerns in law, with human oversight in Article 14 and accuracy in Article 15. ISO 42001 is the management system that makes those duties repeatable and auditable, and it is the certification most translation companies still do not hold.
Our regulated content 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 the EU AI Act (Regulation 2024/1689) and ISO 42001 AI-management governance 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 AI translation for regulated content?
It is a large language model constrained by your approved terminology, followed by a certified human reviewer who takes responsibility for the final text under ISO 18587 and ISO 17100. Raw machine translation on its own does not qualify, because no named person has signed off on the result a regulator will read.
Is AI translation the same as machine translation?
No. Classic neural machine translation works segment by segment with no view of the whole document. AD VERBUM's AI translation is a client-tuned LLM that reads the full document, constrained by your Term Base, then post-edited by a certified linguist.
Does AI translation need human review to be compliant?
Yes. ISO 18587 requires a qualified post-editor who takes responsibility, and MDR Article 10(11) and EMA submission rules hold you to the reviewed text, not the model output. The documented human sign-off is what forms the audit trail.
Is "AI translation vs human translation" the right question?
Not for regulated content. The real choice is AI translation with certified human review versus unreviewed output. AD VERBUM pairs a constrained LLM with certified SME review so you get the speed of the model and the accountability a notified body or the EMA expects.
Where is the data processed?
On EU-hosted infrastructure, with no public-cloud APIs and no training on client data, under ISO 27001. That keeps a dossier or clinical file inside a controlled environment rather than transiting a third-party model service.
What does ISO 42001 add?
ISO 42001 is the AI-management system standard that maps to EU AI Act duties such as human oversight in Article 14 and accuracy in Article 15. AD VERBUM is ISO 42001 certified, which most translation companies are not, so the AI workflow itself is governed and auditable.
