FCA authorised · FRN 724952 0117 325 0027Quote & buy →
Aipedia: Translators and interpreters
AI translation is less reliable in less common and community languages, and the errors are harder for clients to spot. That matters because these are often the languages used in health, legal and public service settings where mistakes cause real harm.
Part of: AI in Translators and interpreters
In short
CIOL has warned since its 2023 AI Voices paper that AI errors are more frequent in less common languages, and its 2025 Translators Day speech said LLMs perform worse in less-resourced languages, including many UK community languages. These languages are often needed in healthcare, policing, courts and immigration. Clients rarely have anyone who can check the output, so errors go unnoticed until they cause harm. A translator or interpreter who relies on weak AI output in these languages is more exposed to a negligence claim. Translators' PI usually responds, subject to the wording, but good practice is to limit AI use in these languages.
Machine translation and large language models learn from large volumes of text. Languages with less published, digitised text, and dialects that differ from the written standard, tend to get less reliable output. Speech tools face the same problem with accents and dialects. The result is output that may be fluent in English but wrong, or output in the target language that reads oddly to a native speaker.
CIOL's AI Voices white paper says AI errors are more frequent in less common languages. In its 2025 Translators Day speech, CIOL said LLMs can amplify bias and perform worse in less-resourced languages, including many UK community languages, and that it would highlight the risks of AI in professional settings to government.
Community languages are often the languages of people using public services: patients, people in police custody, witnesses, asylum applicants and parents dealing with schools. NHS England's 2025 improvement framework raised patient-safety concerns about AI translation apps used across the NHS (Digital Health, 3 June 2025). NRPSI's 2026 article calls for human oversight in high-stakes settings and cites ISO 13611:2014, which favours trained human interpreters for complex, interactive and culturally sensitive communication.
The duty of reasonable care and skill under section 13 of the Supply of Goods and Services Act 1982 does not change with the language, but what is reasonable does. A competent linguist who knows a language pair is poorly served by MT would be expected to rely on it less. In court and tribunal work, the Ministry of Justice expects accuracy, impartiality and confidentiality whatever the language.
Illustrative scenario: a public body asks a freelancer to "check" a machine translation of a safeguarding letter into a community language, at a reduced rate. The engine used a formal register the family would not use, mistranslated a key term about a meeting being compulsory, and rendered a date in the wrong format. The freelancer corrects the date but not the rest. The family does not attend, and the body later complains that the translation was unfit.
Problems that come up repeatedly in these languages include:
A claim alleging that a negligent translation caused loss is usually within translators' PI whatever the language, subject to the wording. Where a translator used AI output in a language pair known to perform poorly, an insurer or court may look closely at whether the review was adequate. Public sector frameworks often set PI as a contract term; check the requirement. Our page for new translators and interpreters covers starting out with cover.
Practical steps for linguists working in less common and community languages include:
Related guides: AI in medical and patient information translation, AI in court and police interpreting and AI translation of immigration and asylum documents.
Apex arranges PI for community and public service linguists. A broker checks your cover against framework requirements and how your policy treats AI-assisted work. Return to the AI in translation and interpreting hub.
CIOL has not issued a ban. It has warned that AI errors are more frequent in less common languages and that LLMs perform worse in less-resourced languages, including many UK community languages, and it stresses human oversight in high-stakes work.
Only if you have tested the engine in that language pair and are confident the output is a sound starting point. Otherwise, price and treat the job as a full translation, and say so in writing so the client understands why.
Usually not in principle: PI usually responds to negligent professional work whatever the language. In practice, a claim may turn on whether your review of AI output was reasonable given known weaknesses, so records of your workflow and decisions help.
Tell us about your firm and the work you do. A broker will review how your current PI treats AI-assisted work, or arrange new cover. No fee for a quote. Or call 0117 325 0027.
Start a PI quote Request a callbackApex Insurance Brokers Limited is authorised and regulated by the Financial Conduct Authority. Registered in England and Wales, company number 07014570. This page is general information about how AI is used in the profession and how insurance may respond. It is not legal, regulatory or insurance advice on your circumstances, and it does not guarantee that cover will be available or on what terms.
