Key takeaways
- UK judicial guidance says information entered into a public AI chatbot "should be seen as being published to all the world".
- Machine translation output can read fluently while getting a term of art, a negation or a number wrong, which is hard to spot without the source language.
- For witness statements filed in civil courts, the translator must certify the translation as accurate, and that responsibility cannot be passed to a tool.
- The Civil Justice Council has consulted on whether rules should address machine translation of court documents, with a final report anticipated later in 2026.
This article is for lawyers, in-house teams, HR and compliance staff, and individuals tempted to run a contract, letter or court document through DeepL, Google Translate or an AI chatbot. It explains what can go wrong, what UK courts and regulators have said, and how machine translation can be used safely within a controlled legal workflow.
For business content outside the legal field, see whether DeepL or ChatGPT is good enough for business translation. For legal translation generally, see our legal translation guide.
What is machine translation, and how is it different from AI chatbots?
Machine translation (MT) is software that translates text automatically, usually using neural networks trained on large volumes of existing translations. Dedicated tools such as DeepL and Google Translate do only this. General AI chatbots built on large language models (LLMs) can also translate, but they generate text by predicting likely word sequences.
Both kinds of tool produce fluent output. Neither understands the legal effect of the text, and neither takes responsibility for accuracy. The difference matters mainly for confidentiality and for how errors appear, not for whether checking is needed.
Why is machine translation risky for legal documents?
Machine translation is risky for legal documents because small errors can change rights and obligations, and fluent output hides those errors from readers who do not know the source language. Typical failures include:
- Terms of art translated as everyday words. "Consideration", "without prejudice" or "deed" may become "reflection", "without harm" or "act".
- Dropped or reversed negations. A missing "not" or "unless" reverses a clause.
- Inconsistent terminology. The same defined term can be translated differently in different paragraphs.
- Numbers, dates and names. Date formats, decimal separators and transliterated names can be silently changed.
- Omissions. Whole sentences can be skipped, especially in long or badly scanned documents.
UK judicial guidance on AI warns: "Public AI chatbots do not provide answers from authoritative databases." It adds that AI tools may "make factual errors" (Courts and Tribunals Judiciary, AI guidance). Our article on legal terms that do not translate shows why terms of art are especially vulnerable.
Is it a confidentiality breach to paste legal documents into a free translator?
It can be. Some free online tools may store or reuse what users enter, and lawyers owe a duty of confidentiality to clients. The same judicial guidance gives a clear instruction:
"Do not enter any information into a public AI chatbot that is not already in the public domain. Do not enter information which is private or confidential. Any information that you input into a public AI chatbot should be seen as being published to all the world."
The guidance is written for judges, but the reasoning applies to anyone handling client documents. If a tool sends personal data to servers outside the UK, it may also be a "restricted transfer" under the ICO's international transfer rules. Our article on confidentiality in legal translation explains the contract terms to put in place.
What do UK court rules say about machine-translated documents?
UK civil court rules make the translator responsible for accuracy, whatever tools are used. For foreign-language witness statements, paragraph 23.2 of Practice Direction 32 says "the translator must sign the original statement and must certify that the translation is accurate".
The Civil Justice Council is considering whether rules should say more. Its interim report and consultation notes that "the use of AI as a translator is well established" and asks whether rules should permit publicly available machine translation, provided the tool is identified and other parties can check the result. According to the CJC's project page, the final report is anticipated later in 2026. Until rules change, assume a human must stand behind any translation filed at court.
When is machine translation acceptable for legal work?
Machine translation is acceptable for legal work when the output is used only for orientation, or when a qualified legal translator fully checks and corrects it. The table shows a practical way to decide.
| Use | Machine translation alone? | Recommended approach |
|---|---|---|
| Sorting a large set of foreign-language emails for relevance | Acceptable inside a secure review platform | Follow with human summaries or full translation of key documents |
| Understanding the gist of a letter you received | Acceptable if the tool is secure and the content is not confidential | Get a human translation before acting on it |
| Contract to be signed | No | Human legal translation with independent review |
| Witness statement or court filing | No | Human translation with the translator's certificate |
| Documents for an authority, such as a registry or the Home Office | No | Human translation in the format the authority requires |
| Internal policy for staff abroad | Only with full human post-editing | Machine translation post-editing by a legal linguist |
Machine translation post-editing (MTPE) means a professional translator edits machine output until it meets an agreed quality level. Our article on MTPE explains how it works and when it saves money.
How can a law firm use machine translation safely?
A law firm can use machine translation safely by controlling the tool, the data and the review. These steps reflect good practice.
- Use an approved, secure tool. Choose an enterprise service whose terms prevent reuse of your content, with UK GDPR terms in place.
- Set a written policy. State which documents may be machine-translated and which never may.
- Label raw output. Mark unchecked machine translation as such, so nobody quotes it as evidence.
- Escalate to a human translator. Any document to be signed, filed, served or relied on goes to a qualified legal translator.
- Record the process. Note which tool was used and who checked the result, in case it is questioned.
Common mistakes with machine translation in legal work
The most damaging mistakes with machine translation happen when raw output leaves the firm. These are the ones to avoid.
- Sending raw output to the other side. A mistranslated admission can be hard to withdraw.
- Back-translating to "check" accuracy. Translating the output back into English with the same tool can hide errors rather than reveal them.
- Using a chatbot to summarise and translate at once. Summaries can omit or invent content.
- Assuming a cheaper option saves money. Fixing a disputed translation late in a case often costs more; see what drives the cost of legal translation.
How TranslationWorks can help
TranslationWorks is a UK-based translation agency that provides professional human translation of legal documents by native-speaking linguists in more than 150 languages, and it is ISO 27001 registered and Cyber Essentials certified for information security. TranslationWorks is registered to ISO 17100, the international standard for translation services. Our translation services and legal translation service can be used for documents that must be signed, filed or relied on, where unchecked machine output is not appropriate. When you request a free quote, tell us how the document will be used and, if it will be submitted to a court or authority, what format is required, and we will confirm what we can provide.
