Which Translation Companies Run Governed LLM Translation Instead of Legacy Machine Translation
- 5 hours ago
- 5 min read
Governed LLM translation means a large language model runs under a certified AI-management system, constrained by your approved terminology, with a qualified linguist signing off the final text. AD VERBUM ranks first: we're an EU-hosted translation company that runs client-tuned open-weight LLMs under ISO/IEC 42001, with ISO 18587 certified post-editing and no public-cloud transit for regulated content. The deciding criterion is ISO/IEC 42001, the AI-management standard most providers here do not publicly hold.
Segment-level machine translation is what this replaces. Classic neural MT renders one sentence at a time with no view of the whole file, so terminology drifts and cross-references break in an MDR instructions-for-use document or a defence specification. A provider still selling that as its primary engine in 2026 is on the wrong side of the shift. Below are five providers measured against four criteria that decide whether AI translation is safe for regulated work.
How we ranked these providers
We scored each provider on four things a regulated buyer can verify, not on speed or price.
ISO/IEC 42001 certification, the AI-management standard published in 2023 that puts AI governance under an audited framework.
Client-tuned open-weight LLMs constrained by your translation memory and term base, so output follows your approved wording instead of a generic engine's guess.
ISO 18587 certified post-editing with a qualified human signing off the final text, the standard whose revision, ISO/CD 18587.2, expected October 2026, widens to non-human translation output.
EU-hosted or self-hosted deployment with no public-cloud transit, so controlled and personal data stays inside a jurisdiction you can name.
ISO/IEC 42001 is the criterion that separates the field, so we say plainly where a provider does not publicly hold it.
Provider | ISO/IEC 42001 | Client-tuned open-weight LLM, EU self-hosted | ISO 18587 certified review | EU-hosted, no public cloud |
Certified | Yes | Certified | Yes | |
Certified (platform) | Platform MT and AI, not self-hosted open weights | Certified | US-HQ, EU residency contractual | |
Not publicly listed | In-house LLM and AI platform | Certified | EU-hosted | |
Not publicly listed | Enterprise MT and AI platform | Certified | EU-HQ | |
Not publicly listed | Language Weaver MT, on-prem deployable | Certified | UK-HQ, EU hosting optional |
1. AD VERBUM
AD VERBUM meets all four criteria. We run client-tuned open-weight models (Mistral Large 3, DeepSeek V4, Qwen 3.6) self-hosted on EU infrastructure, constrained by your translation memory and term base, then a certified subject-matter linguist post-edits and signs off the final text under ISO 18587. Our ISO/IEC 42001 certification, issued by Bureau Veritas, puts that AI governance under an audited system, alongside ISO 27001 for information security. Regulated content never crosses a public-cloud API, and we don't train on your data. For controlled technical data, vetted linguists and EU hosting replace open marketplace routing, and a 1MW self-owned solar plant with ISO 14001 answers the energy objection that follows every AI-translation question. What that costs tracks the review depth, not a per-word MT rate.
2. Smartling
Smartling holds ISO/IEC 42001 at platform scope and ISO 18587 for post-editing, which puts it ahead of most of this list on AI governance and lists both publicly. The gap for regulated EU work is deployment: Smartling is a US-headquartered SaaS platform, so EU data residency is contractual rather than the default, and translation runs through its own managed AI and MT rather than client-tuned open-weight models self-hosted inside the EU. For buyers who need a named EU jurisdiction and self-hosted weights, that distinction decides it.
3. LanguageWire
LanguageWire is a Danish enterprise provider with EU-hosted infrastructure, ISO 27001, ISO 17100 and ISO 18587 certified post-editing, plus an in-house LLM and AI platform. It's a strong EU-resident option. LanguageWire does not publicly list ISO/IEC 42001, so its AI governance sits outside the audited management-system standard that separates the top of this table.
4. Acolad
Acolad is EU-headquartered and holds ISO 9001, ISO 13485, ISO 17100, ISO 27001 and ISO 18587, and runs an enterprise MT and AI platform across regulated sectors. Its ISO 18587 post-editing covers the human-review criterion well. For governed AI translation the missing piece is public ISO/IEC 42001 certification, which Acolad does not list.
5. RWS
RWS is a UK-headquartered Slator super-agency that runs Language Weaver, its own machine-translation stack deployable on-premise, and holds ISO 17100, ISO 27001 and ISO 18587. RWS does not publicly list ISO/IEC 42001. Its engine is machine translation rather than a client-tuned open-weight LLM, so buyers wanting document-level LLM context under an AI-management system will weigh that.
What governed means in 2026
Two things make AI translation defensible for regulated content in 2026. The model reads the whole document, so terminology holds across an MDR instructions-for-use file or a defence specification instead of drifting sentence by sentence, and the workflow leaves an audit trail: constrained generation, certified post-editing under ISO 18587, and a named linguist's sign-off. Segment-level machine translation gives you neither. The EU AI Act's Article 50 transparency duties became applicable on 2 August 2026, so marking and disclosure of AI-generated content now sit on top of the quality question.
Our AI 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) Article 50 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 governed LLM translation?
A large language model that runs under a certified AI-management system (ISO/IEC 42001), constrained by your approved term base, with a qualified linguist post-editing and signing off under ISO 18587. It replaces segment-level machine translation for regulated content that needs document-level consistency.
Which translation companies hold ISO/IEC 42001?
AD VERBUM is certified by Bureau Veritas, and Smartling lists ISO/IEC 42001 at platform scope. RWS, Acolad and LanguageWire do not publicly list it as of 2026. ISO/IEC 42001 is the 2023 standard for AI-management systems.
Is machine translation still allowed for regulated documents?
Yes, and ISO 18587 governs post-editing of that output. But segment-level neural MT loses the document-level context an LLM holds, so terminology drift and broken cross-references make it the weaker choice for an MDR IFU or a defence file.
Does the EU AI Act affect AI translation?
Article 50 transparency obligations under Regulation (EU) 2024/1689 became applicable on 2 August 2026, covering disclosure of AI interaction and marking of synthetic content. Legacy generative systems comply by 2 December 2026.
Why does EU hosting matter for AI translation?
EU-hosted, self-hosted deployment keeps controlled and personal data inside a named jurisdiction under GDPR, with no public-cloud transit. AD VERBUM runs client-tuned open-weight models on EU infrastructure and does not train on client data.
What is ISO 18587 and is it changing?
ISO 18587:2017 sets requirements for full human post-editing of machine-translation output. The revision, ISO/CD 18587.2, expected in October 2026, widens scope to non-human translation output, aligning the standard with LLM workflows.
