AI Will Not Replace Translators. But It Will Expose Those Who Never Understood Translation in the First Place.

Snežana Mitić | February 2026

A widely cited automation risk assessment from the website willrobotstakemyjob.com places translators and interpreters at roughly 70% risk of being replaced by artificial intelligence. The number is attention-grabbing. It is also misleading.

Not because AI lacks capability. It doesn’t. Neural machine translation has made enormous strides, and large language models can produce remarkably fluent text in dozens of languages. The issue is that fluency is not accuracy. And accuracy, in many professional contexts, is not optional. It is a regulatory, legal, and ethical obligation.

This article is about the documents AI should never translate alone, the industries where a mistranslation carries consequences far beyond embarrassment, and why the real conversation is not about human versus machine, but about governance, risk, and accountability.

What the Automation Statistics Actually Tell Us

The willrobotstakemyjob.com risk score is based on a machine learning model trained on occupational attributes from the U.S. Department of Labor’s O*NET database. It measures how susceptible a job’s core tasks are to computerization. For translators and interpreters, the generated score hovers around 70%.

But here is the critical detail most people skip: the model does not distinguish between translating a restaurant menu and translating a pharmaceutical label that will go on a medicine bottle distributed to millions of patients across the European Union. It treats „translation“ as a single task.

That distinction changes everything.

For general-purpose, low-stakes content, AI translation tools are already good enough. They will continue to improve. Translators who only ever handled simple correspondence or basic marketing copy are right to feel the pressure.

But in regulated industries, medicine, law, banking, insurance, pharmaceuticals, translation is not a commodity. It is a controlled process. And it requires something no algorithm currently possesses: professional judgment rooted in domain expertise and an understanding of what happens when a word is wrong.

The Industries Where AI Translation Is a Liability, Not an Advantage

Medicine and Life Sciences

This is the field where the stakes are highest and the margins for error are smallest.

A 2024 study published in AI-Linguistica analyzed neural machine translation output for WHO public health documents across English, French, and Spanish. Deviation from standardized WHO terminology accounted for 16% of all errors. Inconsistent use of terminology contributed another 9%. General mistranslation topped the list at 19%.

When one in four errors comes from terminology problems alone, this presents a serious risk for pharmaceutical regulatory documentation, which relies heavily on precise, repetitive terminology governed by standards such as the Medical Dictionary for Regulatory Activities (MedDRA) and the European Directorate for the Quality of Medicines (EDQM) Standard Terms.

The European Medicines Agency addressed this directly in its 2024 Reflection Paper on the Use of AI in the Medicinal Product Lifecycle, stating that AI and machine learning applications used for translating medicinal product information documents should be used under close human supervision.

Real-world consequences are not theoretical. A BBC investigation published in 2023 found that translation problems contributed to more than 80 cases of infant fatalities or serious brain injuries in the United Kingdom between 2018 and 2022, where hospital staff relied on online translation tools for communication with non-English-speaking patients.

Documents that should never be entrusted to AI alone in medicine and pharmaceuticals include clinical trial protocols, patient information leaflets, summaries of product characteristics, informed consent forms, adverse event reports, regulatory submissions, drug labeling and packaging text, and medical device instructions for use.

Each of these documents carries regulatory weight. A single error in dosage instructions, a misrendered adverse reaction, or a poorly translated contraindication can result in patient harm, regulatory rejection, or product recall.

Law and Legal Translation

Legal language is inherently precise, jurisdiction-specific, and deliberately constructed. A contract that says „shall“ carries different obligations than one that says „may.“ A power of attorney that mistranslates the scope of authority granted can void the entire instrument.

AI translation tools consistently struggle with legal texts because legal terminology is not universal. The same concept may require entirely different phrasing depending on the legal system. Civil law and common law systems use different frameworks, different terminology, and different logic. A term that has a precise meaning in Serbian law may have no direct equivalent in English law, and vice versa.

Confidentiality is another critical concern. Legal documents often contain sensitive information, contract terms, litigation strategy, settlement figures, personal data subject to GDPR protections. Uploading these documents to cloud-based AI translation platforms raises significant data security and confidentiality risks, something many law firms and corporate legal departments now explicitly prohibit in their vendor agreements.

Documents that should never be entrusted to AI alone in legal translation include court-certified translations, contracts and agreements, powers of attorney, corporate governance documents, patent applications, arbitration and litigation materials, compliance documentation, and regulatory filings.

Banking and Finance

Financial translation requires absolute precision with numbers, regulatory terminology, and compliance language. A misplaced decimal point, a mistranslated financial covenant, or an incorrectly rendered clause in a loan agreement can expose institutions to millions in liability.

Banking documents also operate within strict regulatory frameworks. Basel III requirements, anti-money laundering (AML) directives, and Know Your Customer (KYC) procedures all require translations that adhere to specific regulatory vocabulary. AI tools trained on general corpora rarely capture the specificity needed for these documents.

The EU AI Act, which entered its phased implementation in 2024-2025, classifies AI systems used for credit scoring and financial decision-making as high-risk, imposing strict requirements for transparency, traceability, and human oversight. While the Act does not directly regulate translation AI, the principle is the same: when automated systems impact people’s finances and legal rights, governance matters.

Documents that should never be entrusted to AI alone in banking and finance include audit reports, loan agreements and credit documentation, financial statements and annual reports, regulatory compliance filings, anti-money laundering documentation, prospectuses and investment memoranda, and correspondence with financial regulators.

Insurance

Insurance translation sits at the intersection of legal precision and financial accuracy. Policy language is notoriously complex, with specific terms of art that determine coverage, exclusions, and liability.

A mistranslated exclusion clause can leave an insurer exposed to claims they intended to deny. A poorly rendered benefit description can create obligations that were never part of the original policy design. And in cross-border insurance, where policies must comply with multiple regulatory regimes simultaneously, the cost of error compounds.

As of late 2025, 23 U.S. states and Washington, D.C. had adopted the National Association of Insurance Commissioners’ AI Model Bulletin, which requires insurers to establish governance, documentation, and audit procedures for AI systems. The direction is clear: regulators are demanding transparency and human oversight wherever automated systems affect consumer outcomes.

Documents that should never be entrusted to AI alone in insurance include policy wordings and endorsements, claims documentation, reinsurance treaties, regulatory filings and compliance reports, underwriting guidelines, and customer-facing disclosure documents.

The Real Problem With AI Translation in Regulated Industries

The most dangerous thing about modern AI translation is not that it produces obvious errors. It is that it produces fluent, confident, wrong output.

A machine-translated pharmaceutical document may read smoothly. The grammar may be impeccable. But if the system has substituted a close synonym where a standardized regulatory term was required, if it has dropped a negation, if it has hallucinated a reference, or if it has smoothed over a safety-critical distinction that existed in the source text, the result looks professional and is substantively incorrect.

This is what makes AI translation uniquely risky in regulated environments. The errors are not visible to someone who does not have deep domain expertise in both the subject matter and the target language. A project manager running a document through an AI tool and skimming the output will not catch them. A junior reviewer without specialist training will not catch them. These errors require exactly the kind of expertise that professional, domain-specialized translators provide.

Where AI Does Belong in the Translation Workflow

I am not arguing against technology. At TSI, we use translation memory systems, terminology databases, and computer-assisted translation tools every day. Technology is part of our workflow, and it makes us more efficient.

The question is never whether to use technology. The question is how to use it responsibly.

AI can be a productive component of a controlled translation workflow when it is used for initial drafts of high-volume, lower-risk content that will undergo full human review. It can assist with terminology consistency checks across large document sets. It can accelerate the identification of repetitive segments in technical documentation.

But AI should not be the decision-maker. It should not determine final terminology choices in regulated content. It should not be the last set of eyes on a document that carries legal or clinical consequences. And it should never be deployed without a clear quality assurance process that includes qualified human review by translators who understand the domain, the regulatory framework, and the target audience.

The EMA’s position is the right one: AI under close human supervision. Not AI as a replacement for human judgment.

Choosing the Right Translation Partner Is a Risk Management Decision

When a pharmaceutical company selects a translation provider for its regulatory submissions, that decision is a compliance decision. When a law firm engages a translator for court-certified documents, that decision carries professional liability implications. When a bank commissions translations of its financial reporting, accuracy is an audit requirement.

In each of these cases, the question is not „Can AI do this faster?“ The question is: „Is this workflow auditable? Is it fit for purpose? Is there a qualified human accountable for the final deliverable?“

Twenty years of experience in this profession, from my first simultaneous interpreting engagement to the thousands of pages and hundreds of court-certified translations that followed, have taught me something that no algorithm has learned: translation is not about converting words from one language to another. It is about understanding context, consequence, and responsibility. It is about knowing what can go wrong and building processes to prevent it.

AI can support the process. But it cannot replace informed judgment.

At Translation Services and Interpreting (TSI), we integrate technology into a quality-controlled workflow. We use the right tools for the right purpose. And we maintain what matters most: accuracy, regulatory awareness, accountability, and control.

The goal is never speed at any cost. The goal is reliable, compliant outcomes.

Technology can support delivery. Governance protects quality. And choosing the right partner is part of that decision.

TSI agencija iz Novog Sada vam stoji na raspolaganju sa stručnim prevodiocima za 27 jezika, pažljivo biranim prema specifičnostima i potrebama vašeg projekta.

Pišite nam i otvorite svoj sajt i biznis celom svetu!

Autorka: Snežana Mitić, vlasnica agencije TSI Novi Sad
(Za besplatne konsultacije i ponudu, kontaktirajte nas putem sajta ili društvenih mreža.)