AI Translation Gone Wrong: How Global Businesses Are Facing Legal and PR Nightmares in 2026

Updated · 5 min read
On this pageThree relationships, three contractsClauses between buyer and providerLanguage clauses toward customers and partnersWhat engine providers promiseRecords decide disputesThe PR sideCommon questions

A distributor in Korea relies on a machine-translated version of a supplier's warranty terms. The Korean says the warranty covers "all defects". The English says "defects in materials and workmanship". A dispute follows. The supplier says the English prevails; the distributor says it was given the Korean. The translation vendor says it was asked for "fast, AI-assisted" delivery. Each party points to someone else, and the contracts between them say almost nothing about who bears the risk of a translation error.

This post describes issues to raise with your lawyers when drafting or reviewing contracts. It is not legal advice, and the effect of any clause depends on the governing law and the facts.

Three relationships, three contracts

A translation error that causes loss usually involves three relationships, and each needs its own allocation of risk.

Buyer and translation providerWho may use machine translation, on which content, with what human review; the standard of quality promised; liability caps; confidentiality of material sent to engines.
Company and its customers or partnersWhich language version prevails; whether translated content is for information only; how errors are corrected; what notice is given that text was machine-translated.
Company and its technology providersWhat the engine provider promises (usually very little about accuracy); data use and retention; who is responsible for configuration, glossaries and prompts.

Clauses between buyer and provider

Many translation agreements still assume a human translator. They promise "professional quality" without defining it, and say nothing about machine translation. Points to address:

IssueWhat to settle
Use of machine translationAllowed or not, per content type; disclosure when it is used; which engines are approved
Review levelWhether output is fully post-edited, lightly post-edited or delivered raw; who reviews; any independent second review for high-risk text
Quality standardA definition that can be checked: error categories, severity, and what counts as a critical error
ConfidentialityWhether source material may be sent to third-party engines, and on what data retention terms
LiabilityCaps, exclusions, and whether different caps apply to different tiers of content
CorrectionTime to correct errors once reported, at whose cost
RecordsWhat the provider records about each job: translator, reviewer, tools, source version

The most useful single change is to tie the review level to the content type. A contract that allows machine translation with light review for internal documents but requires full post-editing and independent review for contracts, safety text and regulatory documents puts the risk where it belongs. See a framework for when to use machine translation and when to use people.

Language clauses toward customers and partners

For contracts that exist in two languages, a language precedence clause states which version prevails if they differ. It is common, and useful, but it does not make the other version irrelevant: the party who signed the other language may still argue they relied on it, and consumer-protection rules may limit how such clauses work against consumers; ask counsel. A precedence clause is a backstop, not a substitute for an accurate translation.

If you offer a generative AI service to users in Korea, check the AI Basic Act as well. It has applied since 22 January 2026 and requires operators of generative AI to tell users in advance that the service uses AI and to label AI-generated content; regulators have indicated a one-year grace period before fines.

Weak

Precedence clause only

"In case of discrepancy, the English version prevails."The Korean version is machine-translated and never reviewed. The clause may help in a dispute, but the relationship is already damaged, and the outcome is uncertain.
Stronger

Clause plus process

Same clause, plus a reviewed Korean version and a record of review.Errors are unlikely in the first place, and if one appears, the company can show it took reasonable care.

For websites and help centres, a machine translation notice tells readers that a page was translated automatically and which version is authoritative. It sets expectations for long-tail content. It should not be used to excuse unreviewed translation of terms, prices, safety information or legal notices.

What engine providers promise

Read the terms of the translation engine or AI platform you use. They typically disclaim accuracy and fitness for purpose and limit liability heavily. In practice, the risk of an error stays with the company that published the output. That is why the review and record-keeping clauses in your other contracts matter: they are where you can actually manage the risk.

Check also how the provider may use your data. Sending confidential documents to an engine whose terms allow use of inputs for training can create a separate problem, unrelated to translation quality. See when machine translation exposes confidential data.

Records decide disputes

When a translation error causes a dispute, the first questions are factual: which source version was translated, how, by whom, and who approved it. Companies that cannot answer them are in a weak position whatever their contracts say.

  1. Keep the source version and the translation together, with dates.
  2. Record whether machine translation was used, and which engine and settings.
  3. Record who post-edited, who reviewed, and who approved publication or signature.
  4. Keep correction history: what was changed, when, and why.
  5. For bilingual contracts, keep the negotiation drafts in both languages.

The PR side

A legal dispute about a translation is often also a public one. Customers share screenshots; partners complain to the press. A company that can say "this text was reviewed by a qualified translator and approved by our legal team; we have corrected the error and contacted those affected" is in a different position from one that says "the AI did it". For the communications side, see why human post-editing still protects corporate reputation; for the courtroom side, when AI translation fails in court; and for high-stakes text generally, a four-tier risk model for legal and medical translation.

Common questions

If our translation vendor used AI without telling us, who is responsible for errors?

It depends on your contract and its governing law. Add clauses on whether machine translation may be used and with what review.

Does a disclaimer on machine-translated web pages protect us?

It sets expectations and may help, but do not rely on it for content that must be accurate.

What records should we keep?

Source version, who or what translated it, who reviewed and approved it, and when.

Bilingual contracts and terms

We translate and review Korean-English contracts and terms with full records of who did what. Talk to us →

Work with us

Contracts that read the same on both sides.

Korean legal and compliance translation reviewed for meaning, not just wording.

Get a free quote