The EU AI Act and Translation: Why Human Expertise and Traceability Matter

When a business publishes content in several languages, can it explain how it produced each version? The original might come from a copywriter, the translation from an AI tool, and the final corrections from someone whose role was never documented.

The EU’s Artificial Intelligence Act (EU AI Act) gives businesses fresh reasons to examine these processes. But what do its transparency rules mean for translated content, and where can human expertise help reduce risk?

In this article, we’ll look at the rules affecting translated content and how human expertise can help businesses manage risk. Please note that this is a general overview rather than legal advice; specific compliance questions should be discussed with a qualified legal adviser.

EU AI Act content transparency requirements

Article 50, the part of the EU AI Act that covers transparency obligations, started applying on 2 August 2026. It requires organisations to inform people when they interact with AI, unless it’s “obvious from the point of view of a natural person who is reasonably well-informed, observant and circumspect”, and to disclose deepfakes and AI-generated or manipulated images, audio, or video that falsely appear authentic. However, that doesn’t mean that every AI-translated document needs a disclaimer.

Technical marking and public disclosure

The rules distinguish between organisations that provide AI systems, known as “providers”, and those that use them professionally, known as “deployers”.

Generally, providers of AI systems that generate synthetic text, images, audio or video must ensure their outputs include machine-readable markings that make AI involvement detectable. These technical markings (which can include digital watermarks) differ from a visible disclosure explaining AI use to readers.

When does text need a disclosure?

For deployers, the text disclosure requirement concerns AI-generated or manipulated material published to inform the public on matters of public interest. These can include public health, environmental protection and consumer safety.

However, an exception applies when the content undergoes meaningful human review or editorial control and a person or organisation holds editorial responsibility for publication. Where disclosure is required, it must be clear, distinguishable and accessible by the reader’s first exposure.

Breaches can bring fines of up to €15 million or 3% of worldwide annual turnover for the preceding financial year, generally whichever is higher. Lower ceilings apply to certain smaller businesses, and individual circumstances may influence the penalty imposed.

What counts as meaningful human review?

Human review involves more than checking spelling and grammar. The European Commission’s guidance describes a deliberate examination of the content by someone with relevant knowledge and professional judgement. Editorial control means having the authority to approve, alter or reject the content. A person or organisation must also hold editorial responsibility for publication.

Where machine translation post-editing fits

Machine translation post-editing (MTPE) involves reviewing and editing machine-translated output. Describing content as “AI-assisted” rather than “fully AI-generated” does not, by itself, establish whether disclosure is required. The substance of the review and editorial responsibility remain important.

An editor might correct awkward phrasing, compare every sentence against the source, or resolve specialist terminology with the client. These tasks provide different levels of scrutiny.

Businesses should therefore establish what the review covers and who is responsible. A translator can assess whether the target text accurately reflects the source, while a subject-matter expert may need to validate the information itself.

How human translation supports AI Act compliance

Consider a Japanese manufacturer preparing a consumer-safety notice in English. Readers need to understand which products are affected, whether the instructions apply to their circumstances, and what to do if they are. 

A translation can sound fluent while changing crucial details. A missing condition might make an instruction too broad, while an ambiguous reference could leave customers unsure whether their product is affected.

A professional translator can compare the text with the source, flag uncertainty and ask the manufacturer to clarify its meaning. A second linguist can revise the translation to identify discrepancies before publication.

This is where human expertise provides practical protection: resolving uncertainty before it reaches readers and helping ensure the translated notice communicates what’s intended. It also helps establish when another specialist is needed. A translator can convey a technical claim accurately, but the manufacturer’s subject-matter expert must confirm the claim itself is correct.

The importance of documenting the translation process

If a business relies on human review, keeping a record of that work can help it show how it assessed the content. Useful records might include the source version, people and tools involved, substantive queries, corrections and final approval. These can help establish what was checked and who took responsibility.

Keeping these records is a practical risk-management measure, rather than a specific Article 50 requirement to document every translation project. They can also make it easier to investigate a disputed passage or update translations when the source changes.

What translation standards demonstrate

ISO 17100 sets requirements for translation-service processes and resources, providing a foundation for consistent quality assurance. However, it does not cover raw machine translation followed by post-editing.

Documented quality processes may provide useful evidence when assessing how content was produced and reviewed. However, neither a project record nor translation-service certification guarantees compliance with the EU AI Act. Businesses still need to assess the applicable obligations for the specific workflow and publication.

What should businesses review now?

A useful starting point is to identify where AI is used to create and translate content, including any work handled by external suppliers. Asking a few simple questions can help:

• Does the content need an AI disclosure under Article 50?
• Who checks the translation, and does it need a subject-matter expert?
• Who takes editorial responsibility for publication?
• What records show how the content was reviewed?

These answers can help businesses choose the right approach and clarify responsibilities. Legal advice may be needed where obligations are uncertain, while a language partner can help put the translation and review process into practice.

Human expertise behind every translation

By establishing clear responsibilities and a careful review process, businesses can manage the risks of multilingual publishing. Word Connection supports this work through professional English, French and Japanese translations, revisions and localisation projects, helping organisations communicate accurately across languages and cultures.

To discuss your next translation project, contact our team.