AI has long arrived in law firms and fiduciary offices: for drafts, translations, summaries and research. The benefit is real, because much of the work consists of text. At the same time these professions work with information subject to special secrecy protection that often contains sensitive personal data. This article describes how to make its use responsible.
Professional secrecy first
Lawyers, notaries and auditors bound to confidentiality under the Swiss Code of Obligations are subject to criminally protected professional secrecy. Fiduciaries are bound to confidentiality by contract and, depending on the activity, by law. For AI use this means: if client information is given to an external service, it must be ensured that the service is bound to confidentiality like an auxiliary, learns only what is necessary and does not use the information for its own purposes. The professional-law assessment in each case lies with the firm and, where relevant, the competent professional body.
Data protection on top
Client files often contain sensitive personal data, for example on health, criminal proceedings or financial circumstances. Under the revised FADP, an AI provider is usually a processor; a contract is needed, and for processing outside Switzerland and the EU a basis for disclosure abroad. Where risk is high, a data protection impact assessment is advisable. The rules for disclosures abroad are described in our article on data transfers.
The right tools
The most important difference lies in the account type. With consumer accounts of many AI tools, inputs may be used to improve models; with business access this is often excluded by contract. For client information, therefore, only tools with a company account, a contract, training excluded, a known processing location and a clear retention period qualify. Equally important: review AI features newly appearing in existing practice or accounting software before use.
Minimise data
Many tasks need no identifying details. A contract draft can be improved with placeholders instead of names; a summary of public judgments needs no client data at all. Making a habit of removing names, addresses and details before input lowers the risk considerably without giving up the benefit.
Check every citation
Language models phrase things convincingly even when wrong. In several countries courts have reprimanded filings containing invented judgments and citations from AI tools. Law firms therefore apply a simple rule: AI outputs are drafts. Judgments, statutory provisions and citations are checked against the primary source, and deadlines are never taken from an AI output.
Transparency towards clients
Clients expect their matters to be handled confidentially and carefully. A short provision in the terms of engagement on how AI is used and which safeguards apply creates clarity. A chatbot on the website must be labelled as AI where the AI Act applies and should not give professional answers; the trade-off is covered in our decision guide AI chatbot or website search.
Organisation rather than bans
A ban on AI tools gets bypassed in practice. More effective is a short framework: approved tools, data rules with examples, a duty to check, an incident reporting route and training for everyone, including assistants and trainees. The steps are detailed in our checklist on AI in law firms; building a general policy in the AI usage policy checklist.
Where AI genuinely helps
With drafts and rewording, summaries of public sources, translations and search across your own templates and documents. For the last point, an internal assistant that accesses only your own documents and cites the source for every answer pays off; that is described in our internal AI assistant with source citations offer. For introducing a framework with tool review and training there is the AI usage policy and vendor review package.