AI translation under the MDR: is it permitted?

Manufacturers, importers, distributors, PRRCs and quality/regulatory teams keep asking the same question: is AI translation permitted under Regulation (EU) 2017/745 (MDR)?

The short answer is yes: AI tools may be used for translation, but their use is subject to certain conditions and restrictions. In practice, some auditors or bodies expect evidence of process control, which can give the impression that “AI is prohibited” when, in fact, the real issue is a lack of control, traceability and validation.

1) The theoretical framework: which texts govern the translation of medical devices?

MDR (EU) 2017/745: language and information requirements, not translation methods

The MDR sets stringent requirements for the content and quality of the information provided with the device, including labelling and instructions for use. In particular, it emphasises that instructions and information must be written in terms that are readily understood by the intended user. In other words, regardless of the tool used, the final text must be clear, consistent, accurate and safe.

The MDR also sets out language requirements: the information provided with the device must be available in one or more of the official languages required by the Member State where the device is made available. These requirements are often determined by national measures, which is why a country-by-country approach is justified.

Key point for AI translation: the MDR does not state a specific translation method. What it does say, however, is that you must ensure that the information provided is “complete and correct” and that you must be able to demonstrate this through a quality management system.

The AI Act: not a ban, but a framework of risks and obligations

The AI Act establishes a horizontal framework covering the definition of AI systems, obligations for the parties involved, and stricter requirements for certain high-risk systems, among other things. Like the MDR, the AI Act does not prohibit the use of AI, but emphasises that it must be subject to appropriate controls, particularly in terms of governance, human oversight, confidentiality and data management.

2) What does the MDR say about the use of AI translation?

Within the scope of medical device translation, the MDR makes no mention of AI, nor does it state that “human translation is mandatory” or that “machine translation is prohibited”. Instead, it sets out requirements that have a direct impact on translation: the information must be easily understood by the intended user; the labelling and instructions for use (IFU) must contain specific mandatory information; and compliance must be demonstrable through a quality management system and controlled documentation.

So the real question is: can you demonstrate that the final translation is reliable and subject to appropriate controls?

Why the “perceived ban” comes up during audits

Some auditors ask for evidence that the translation was not “generated using AI”. This type of request may stem from poor wording (the auditor is actually concerned about an uncontrolled or overly informal process), heightened sensitivity to patient risk, or a need for traceability (source and target versions, change management and validation). In a previous article, published in October 2025, on the use of AI in translation, we discussed the “Who?” and the “How?”.

Who is the person using AI? Is it a professional translator or someone without the necessary qualifications for the task?

How? In other words, which tool is being used? Is it a publicly available LLM (Large Language Model) or a tool specifically designed and adapted for translation?

The outcome depends on the “Who?” and the “How?”, and that is what the auditor is really interested in.

3) Practical steps for ensuring the safe use of AI in translation

In an MDR environment, the issue is not the use of AI itself, but the ability to manage its risks and demonstrate process compliance.

To ensure that medical translation complies with regulatory requirements and meets the expectations of notified bodies, several key measures need to be put in place:

• Suitable tools: use professional AI solutions that ensure confidentiality, terminology consistency and traceability
• Human responsibility: ensure that the translation is carried out or supervised by a qualified professional with a sound understanding of the MDR/IVDR context
• Systematic review: ensure that every AI-generated translation is reviewed, corrected and validated by a human
• Documented validation: incorporate final approval into the quality management system (ISO 13485), with supporting evidence
• Terminology control: use validated glossaries for critical content (IFUs, labelling, safety information)
• Full traceability: retain records of versions, approvals and tools used to meet audit requirements

In practice, combining these measures enables AI to be used in compliance with regulatory requirements, while mitigating regulatory risks and ensuring the reliability of translation processes.

4) How can you ensure the compliant use of AI?

If an auditor explicitly requests that you “prove that this translation was not done using AI”, the discussion should be redirected away from the tool itself and towards process control and the associated evidence.

The aim is to demonstrate that the translation is controlled, validated and compliant, regardless of the method used.

To achieve this, you can draw on:

• Documented procedures: a clear description of the translation and review process, including roles, responsibilities and acceptance criteria
• Evidence of qualifications: the qualifications of those involved and the selection of service providers
• Validation records: checklists, corrections, formal approvals and revision histories
• Document management: version control, source/target alignment and change tracking
• Justification of controls: demonstrating that critical content (warnings, safety, IFUs) is subject to enhanced scrutiny
• AI governance (if used): description of the tool used, supervision measures in place, restrictions on use and data management

In an audit, the expectation is not that AI has not been used, but that you can demonstrate that its use is appropriately managed and controlled and does not affect compliance.

5) Conclusion

AI translation is not prohibited under the MDR (EU) 2017/745. The regulation focuses primarily on the quality of the information supplied and the processes used to manage it. It is not prohibited under the AI Act either. It reinforces the need to ensure that tools and processes are properly controlled, particularly through governance and risk management.

To avoid non-conformities arising from the use of AI, whether for translation or any other purpose related to medical devices, it is essential to describe and document the workflow, the tool used, the governance arrangements and the validation process.

Ultimately, AI must be treated as a support tool within a controlled quality process.

If you outsource your translations, a translation company specialising in the medical device sector can help you establish a controlled workflow (review, terminology, evidence and confidentiality) while ensuring that deadlines and language requirements are met.