Machine translation, AI and professional secrecy

Read the PDF edition (3 pages)

Machine translation and generative artificial intelligence have changed the practice of multilingual drafting within a few years. Neural translation systems produce translations of legal texts that are often fluent and mostly accurate. Language models can compare versions, suggest terminology and draft clauses in several languages at once. Law firms, banks and pharmaceutical groups in Basel use such tools for first drafts, for checking consistency between versions and for understanding documents in languages their lawyers do not read. The tools save time and reduce cost, yet they introduce new risks that a careful practitioner must manage.¹

The first risk is accuracy. Machine translation systems reproduce the most common equivalent of a term in general usage, which is often the wrong legal term. They may render Gewährleistung as warranty, Garantie as guarantee and Treu und Glauben as good faith without warning the user of the conceptual differences discussed above. They may also omit a negation, confuse obligations with permissions or translate a defined term inconsistently across a long document. The output must therefore be reviewed by a lawyer who knows both languages and the governing law. That review should be documented when the translation is used in a transaction.²

The second risk concerns confidentiality. A lawyer in Switzerland who enters client information into an external translation or AI service discloses it to the provider of that service. Article 321 of the Swiss Criminal Code punishes the disclosure of secrets entrusted to lawyers in their professional capacity, while Article 13 of the Federal Act on the Free Movement of Lawyers imposes a professional duty of confidentiality. German lawyers are bound by section 203 of the Criminal Code and by section 43e of the Federal Lawyers' Act, which regulates the use of external service providers. Swiss lawyers who wish to use cloud-based tools must therefore ensure that the provider is contractually bound to confidentiality, that access by third parties is excluded as far as possible and that the client has consented where the risk requires it.³

Data protection law adds a further layer. The revised Federal Act on Data Protection, in force since 1 September 2023, permits the processing of personal data by processors under Article 9 if the controller ensures that the data are processed only as the controller itself would be permitted to process them. Under Article 16, the disclosure of personal data abroad is permitted where the Federal Council has found that the destination state ensures adequate protection, or where appropriate safeguards are provided. Banks are additionally bound by banking secrecy under Article 47 of the Banking Act and by the requirements of the Swiss Financial Market Supervisory Authority on outsourcing. For many transactions in Basel, the practical solution is to use translation and AI tools that run in a controlled environment, to anonymise documents before processing them or to restrict the tools to texts that contain no confidential information.⁴

Terminology management is the constructive answer to many of these risks. A firm or legal department that drafts contracts in several languages benefits from a glossary that records the agreed equivalents of key terms in each language and legal system, together with notes on the differences. Such a glossary may state that Garantie in the sense of Article 111 of the Code of Obligations is rendered as independent guarantee, with guarantee alone reserved for a different concept, or that representations and warranties is rendered as Zusicherungen und Gewährleistungen only with an express exclusion of the statutory regime. Translation memory systems can enforce such glossaries automatically, and language models can be instructed to follow them. A glossary of this kind is part of the institutional memory of a legal team and deserves the same care as its precedents.⁵

Notes

1. On the use of machine translation and generative AI in legal practice, see the guidance published by bar associations and the professional literature since 2023. Status to be verified. ↩
2. On typical errors of machine translation in legal texts, see studies in translation studies and legal linguistics. Specific studies to be verified. ↩
3. Art. 321 of the Swiss Criminal Code (StGB, SR 311.0); Art. 13 of the Federal Act on the Free Movement of Lawyers (BGFA, SR 935.61); § 203 of the German Criminal Code (StGB); § 43e of the Federal Lawyers' Act (BRAO). To be verified. ↩
4. Federal Act on Data Protection (FADP, SR 235.1), in force since 1 September 2023, Arts. 9 and 16; Art. 47 of the Banking Act (BankA, SR 952.0); FINMA Circular 2018/3 on outsourcing. Status to be verified. ↩
5. On terminology management in legal translation, see the literature on legal terminology databases, for example the federal terminology database TERMDAT of the Swiss Federal Chancellery and the European Union's IATE database. To be verified. ↩

For general information only. This article does not constitute legal or tax advice, does not replace advice on an individual case and creates no lawyer-client relationship. It reflects the legal position as of August 2025.