Artificial intelligence promises speed and lower costs in processes that used to require hours of specialized work. In technical and legal translation, that promise appeals to companies and professionals who need fast turnaround. The problem appears when the result reaches the court, the consulate or the contract: distorted terms, invented case law and clauses with no legal equivalent in the target language. Cases of lawyers fined for citing court decisions that never existed show that the risk is real and the consequences are costly.
Key takeaways
- AI tools invent nonexistent citations and case law, a technical phenomenon called hallucination, with documented legal consequences in Brazil
- The Santa Catarina State Court of Justice fined a lawyer who used fake AI-generated case law in an appeal
- The Regional Labor Court of Minas Gerais (TRT-MG) upheld a fine after a lawyer cited a nonexistent precedent (súmula) created by artificial intelligence to challenge a medical report
- Automated tools take no legal responsibility for errors: the responsibility falls entirely on whoever signed the document
- AI does not understand cultural nuances or legal hierarchies across different legal systems, producing translations that look correct but change the meaning of clauses
- Brazilian sworn public translators and technical translators are civilly and criminally liable for the content they sign, creating a level of accountability that no automated system offers
Why AI fails at legal and technical translation
Hallucinations: when AI invents what does not exist
Generative AI tools can produce fluent sentences that are not in the original document. In legal texts, this shows up as an added clause, a wrong statute number, a changed name or date, or a passage left out without warning. Because the final text looks right, the error is usually found only when the authority or the other party questions the document.
Different legal systems: what AI does not see
Legal systems vary profoundly between countries. A term with a precise meaning in Brazilian law may have no direct equivalent in Italian, German or US law. AI translates word for word or phrase by phrase, without understanding that the hierarchy of norms, procedural concepts and contractual implications work differently in each legal system.
Technical and sworn translators know these differences and choose the correct functional equivalent, not the literal one. This distinction determines whether a contract is enforceable or void in the destination country.
No legal accountability
Sworn translators take civil and criminal responsibility for the content they sign. When an error causes harm, there is an identified, registered professional who can be held accountable. AI tools take no responsibility at all. The error falls on whoever used the tool and signed the document, whether that is the lawyer, the company or the professional who hired the service.
Cases that show the risk in practice
TRT-MG: a nonexistent precedent used to challenge a medical report
The Regional Labor Court of Minas Gerais upheld a fine against a worker after his lawyer cited a nonexistent precedent (súmula) generated by artificial intelligence. The false text was submitted to challenge a court-appointed medical expert’s report. Appellate judge Weber Leite de Magalhães Pinto Filho, the rapporteur, stated that this was not a simple mistake but the creation of nonexistent content with the potential to mislead the court. The judge stressed that using AI tools does not remove responsibility for what is submitted. The full case is available on the TRT-MG website.
TJSC: nonexistent decisions, legal scholarship and judge in the same petition
The Santa Catarina State Court of Justice imposed a fine of R$3,700 after a lawyer filed an initial petition with decisions, a scholarly citation and a judge’s name that did not exist. Judge Daniel Carvalho Martins, of the Concórdia Labor Court, noted that the elements found went beyond a clerical error and reinforced the argument that the petition had been produced by generative AI without human review. The judge explained that language models can “hallucinate” answers, the technical term for generating false information that looks true. The full case is on the TRT-12 website.
What do these cases have to do with translation?
Both cases involve legal filings, but the failure mechanism is the same one that occurs in AI-generated translations: the system produces plausible content without checking whether it is correct. In a legal translation, the equivalent is a procedural term translated in a technically wrong way, a contract clause whose meaning is reversed in the target language or a legal reference with no counterpart in the receiving country’s legal system.
The only difference between a legal filing with fake case law and a contract with a poorly translated clause is the type of document. The legal and financial risk for whoever signs it is the same.
When AI can and cannot be used in translation
Technology has a legitimate role in the translation process. CAT (computer-assisted translation) tools boost human translators’ productivity, identify repetitions and apply glossaries consistently. Project management platforms automate assignment and tracking. These applications work because human judgment remains at the center of the process.
The problem starts when AI replaces the translator instead of supporting them. Legal, technical and certified documents require professional responsibility for the content. That responsibility only exists when a qualified translator has read, interpreted and signed the translation.
How Zetta can help
At Zetta, technology is part of the process to make translators more efficient, not to replace them. Every document goes through qualified public or technical translators, who are accountable for the content and ensure the legal validity of the delivery. Visit our platform to compare translators by specialization and turnaround time, or talk to our team for guidance on your project.