On the 18th of March, the European Commission put forward EU Inc., an optional company form intended to make it easier for businesses to set up and operate across the Single Market. The main idea is to strip some of the friction out of cross-border administration by providing companies with a common European framework rather than leaving every filing, document and procedural step to be dealt with afresh in each member state.
Translation is one of the practical obstacles the proposal tries to reduce. Article 33 begins with a surprisingly direct principle: translation “shall not be required” where the necessary company information is already available through specified records, certificates or interconnected registers. Where a certified translation is still justified and strictly necessary, however, the proposal goes further. Member States could allow that requirement to be satisfied by an approved AI translation agent, including for some administrative and judicial purposes.
The question being addressed therefore has shifted considerably. It’s not just about how fast or smooth a translation is anymore. A certified translation is something people are meant to trust, and that trust usually come from real, identifiable translators, set procedures, and work that can be checked or challenged if something goes awry.
EU Inc. is still only a proposal, and Article 33 applies to a limited category of company documents. Despite that narrow scope, though, it suggests an important question for translators and cross-border specialists:
What should an AI system have to prove before its translation is allowed to carry credible weight?
The Most Overlooked Part of Article 33 Is What It Removes
The first translation reform in Article 33 is administrative rather than technological. EU Inc. is designed so that core company information can already move across borders through recognised certificates and interconnected registers. If an authority can retrieve what it needs from those sources, asking the company to commission another certified translation risks duplicating information that is already available.
That is part of the burden the Commission is trying to reduce. Its wider 28th Regime proposals refer to the “high costs and delays” created by mandatory certified translations for startups and scaleups operating across borders.
There is a strong case for simplifying that process. Where the same authoritative company information is already available in a form another Member State can use, a fresh certified translation may add little beyond another procedural step.
Article 33 becomes more consequential once that reasoning runs out. If a certified translation is still considered necessary, the proposal does not remove the requirement. It changes the way it may be fulfilled.
That is the distinction the rest of the article turns on. Cutting redundant translation is one reform. Allowing AI to satisfy a translation requirement that still carries official weight is another.
One EU Rule, Many Standards of Trust
Europe’s multilingual system is not a small administrative detail. With so many language combinations in play, the idea of an AI translation agent being broadly “approved” becomes much harder to define. A system may perform convincingly in one language pair and far less reliably in another, especially once legal terminology and national company law enter the equation. Approval therefore cannot sensibly mean that an agent is simply good at translation in the abstract.
Source: European Parliament, Multilingualism in the European Parliament
That complexity is compounded by the fact that certified translation is not governed by one uniform European system. The EU’s own e-Justice material describes national arrangements as “heterogeneous”. Some countries rely on sworn translators appointed or recognised by public authorities, while others use different forms of authorisation, certification or professional oversight. Article 33 would leave another important decision at national level too. Each Member State would choose which AI translation agents it is prepared to accept for its own administrative and judicial procedures.
That makes the word “approved” do a great deal of work. A tool that performs well for one language pair or one type of company document has not automatically proved itself across hundreds of other combinations, or in a more consequential legal setting.
A credible approval regime would need to define its limits. A system tested for French-to-English company certificates has not necessarily demonstrated the same reliability in another language pair or on a different class of document. Nor can approval be treated as permanent if the model, terminology resources or configuration later change.
If an official translation is disputed, someone needs to be able to establish which system produced it, which version was running, what linguistic resources were applied and whether the output came from the same configuration that was originally approved.
EU Inc. may create a common corporate framework, but Article 33 could still leave Member States with different standards for when an AI system has earned enough trust to carry official weight.
Who Would Actually Use an Article 33 Translation Agent?
The potential audience for Article 33 is not insignificant. The Commission expects EU Inc. to appeal particularly to smaller companies looking beyond their domestic market, where repeated filing and translation requirements can quickly turn into another cost of expansion.
Earlier Commission research points in the same direction. In a consultation of more than 2,800 SMEs, 46% said they planned cross-border business for which translation would be useful, while more than 70% regarded automated translation as useful to their business. The research predates EU Inc. and cannot tell us how many firms would rely on AI for a certified translation, but it does show that language is already part of the practical burden of moving into another market.
Article 33 is at its most persuasive when that burden involves highly structured company information. Certificates, register material and standard corporate filings contain recurring fields and relatively predictable terminology. In that setting, a tightly controlled automated system has a clearer task than it would with a heavily negotiated agreement or a document whose meaning turns on individual legal phrasing.
The difficulty comes when the same administrative document acquires a different purpose. Article 33 expressly allows for certified translations to remain necessary in judicial proceedings. A company certificate filed routinely with an authority and the same certificate later relied upon in contested litigation may contain identical words, but the translation is no longer doing the same job.
That is an important limit to the efficiency argument. Article 33 may prove most useful precisely where the language is structured and the purpose narrow. The further a translation moves from routine administration towards interpretation, dispute or legal consequence, the stronger the case becomes for retaining identifiable professional judgement alongside the technology.
The Stakes Change When the Translation Enters a Legal Process
Article 33 becomes harder to judge once the translated document is no longer just passing between registers and authorities. The proposal also allows for certified translations where EU Inc. documents are needed in judicial proceedings. At that point, the cost of getting something wrong changes.
A company certificate or set of articles may contain facts about legal form, representation, ownership or corporate authority. In routine administration, an imperfect rendering may create confusion or delay. In a dispute, the same wording may be examined because somebody is relying on it. A mistranslated qualification, an unclear corporate title or a term given the wrong legal sense can become something the parties have to argue about rather than simply correct.
That is where an identifiable professional still offers something that approval alone cannot easily reproduce. If wording is challenged, there is a person or provider who can return to the source, explain the decision and, where necessary, revise it. With an automated translation, the route back is less straightforward.
There is an irony here. Article 33 is meant to remove friction from cross-border business. If uncertainty over an AI-produced certified translation later must be resolved during proceedings, some of that friction simply reappears in a more expensive place.
The question is therefore not whether AI can translate the document. It is whether the chosen process remains dependable when somebody has a reason to challenge what the translation says.
The Question Is What Approval Really Means
EU Inc. has a clear strength where it removes translation that serves no practical purpose. If the same company information is already available through recognised multilingual records, asking for another certified version adds little.
The more difficult part of Article 33 is what happens when a certified translation is still required. In those cases, Member States would be deciding when an AI system is good enough to take on work that carries legal or administrative weight. That may be entirely reasonable for standardised company material, but the judgement cannot be made in the abstract. The document, language pair and intended use all matter.
Article 33 will therefore be judged less by how easily AI can produce a translation than by how carefully those limits are set. If the system is going to be trusted officially, there must be a clear basis for that trust, and a clear answer when the translation is questioned.
For support with certified translations on your next project, submit your details below and the Aplomb team will be in touch.
