explainerAI Governance

Must lawyers tell clients, or ask their consent, before using AI?

Italy legislates a duty to inform, the Netherlands asks for consent, Belgium and Germany say there is no duty: 76 reviewed country answers on telling clients.

Edited and verified by SafeLegalAI

Researched with AI assistance · sources verified by SafeLegalAI · How this was made ↓

Only a handful of countries make it a rule. As of 5 October 2026, the client consent and transparency category records four binding answers (Finland, Italy, Poland and Slovakia) and 26 guidance answers, including ABA Formal Opinion 512 in the United States. The Netherlands asks for client consent; Belgium and Germany say no duty to tell exists.

How was the count made?

The count reads the client-consent-transparency cell of each country record in src/content/countries/ that is not provisional: 76 records. Four are binding, 26 are guidance, 39 record none and seven are unclear. The 54 provisional records sit outside the count; 53 of them record none, and Kazakhstan’s provisional cell is binding under its AI Law.

The category asks two separate questions. Must a client be told that AI is used? Must the client agree before it is used? The documents answer them differently, and some answer only one.

Where it is binding

The binding answers are Finland, Italy, Poland and Slovakia. Binding here does not mean a general duty to announce AI use.

Italy is the only statute in the group. Law 132/2025, Article 13(2), requires every intellectual profession, lawyers included, to tell the client about the AI systems it uses. Slovakia’s bar resolution makes express informed consent one of three permitted routes for putting confidential information into an AI tool. Finland’s bar data-security rules allow non-business-grade services only with the client’s consent. Poland’s amended advocates’ ethics code goes the other way: no general disclosure, but a client may ask which tools are used and why.

Where it is guidance

Guidance answers are Argentina, Australia, Belgium, Brazil, Canada, Czechia, Denmark, the European Union, France, Germany, Indonesia, International bodies, Ireland, Israel, Japan, Malaysia, the Netherlands, New Zealand, Nigeria, Singapore, Spain, Sweden, Switzerland, Türkiye, the United Kingdom and the United States.

Inside that group the documents take three positions. A consent position: the Dutch NOvA recommendations, the Czech Bar opinion, the Israel Bar opinion At/60/24 for client personal data in open systems, and ABA Formal Opinion 512 for confidential information entered into a self-learning tool. An inform position: Brazil’s OAB Recommendation 001/2024 wants written notice before use, and the Bar Standards Board guidance for barristers in England and Wales requires transparency where AI “materially impacts the nature or scope” of the service (rule rC19). A no-duty position: Belgium’s OVB and Avocats.be guidelines and the German BRAK guide.

In Canada, the law societies of Ontario and British Columbia both leave disclosure to judgment. The Law Society of Ontario white paper lists factors for deciding whether to disclose, including whether the client reasonably expects a licensee to have prepared the work.

Where nothing exists

The 39 none records are Austria, Bahrain, Bangladesh, Chile, China, Colombia, Costa Rica, Croatia, Cyprus, the Dominican Republic, Ecuador, Egypt, Ghana, Greece, Iceland, India, Kenya, Kuwait, Latvia, Lithuania, Luxembourg, Malta, Mexico, Morocco, Nepal, Pakistan, Panama, Peru, the Philippines, Portugal, Qatar, Saudi Arabia, Slovenia, South Africa, South Korea, Sri Lanka, Taiwan, the United Arab Emirates and Uruguay.

The seven unclear records are Estonia, Hong Kong SAR, Hungary, Norway, Romania, Thailand and Vietnam. None means the category was checked and nothing was found; unclear means the recorded sources do not yet support an answer.

What the documents actually say

Four passages mark the range, from a statute to an express statement that no duty exists.

Italy, Law 132/2025, Article 13(2): “le informazioni relative ai sistemi di intelligenza artificiale utilizzati dal professionista sono comunicate al soggetto destinatario della prestazione intellettuale con linguaggio chiaro, semplice ed esaustivo.” In our translation, information on the AI systems the professional uses is communicated to the client in clear, simple and complete language.

The NOvA recommendations, under confidentiality, point 2: “Vraag toestemming aan de cliënt voor AI-gebruik in het dossier.” In our translation: ask the client’s permission for AI use on the file.

The OVB and Avocats.be guidelines, section 1.1: “De advocaat is niet verplicht aan de cliënt mee te delen dat beroep wordt gedaan op AI, net zomin als hij wordt geacht te melden dat hij andere IT-toepassingen gebruikt.” In our translation, a lawyer need not tell the client that AI is used, any more than for other IT applications.

ABA Formal Opinion 512, part B, page 7, on consent before client information goes into a self-learning tool: “merely adding general, boiler-plate provisions to engagement letters purporting to authorize the lawyer to use GAI is not sufficient.” Part C adds that disclosure under Model Rule 1.4 depends on the “facts of each case”.

What it does not cover

This category does not decide whether client data may enter an AI tool at all; that is the client-data category, explained in Can lawyers put client data into AI tools?. Telling a court about AI use is the disclosure-in-filings category. Billing for AI-assisted work is the fees category.

The EU AI Act sits beside these rules without replacing them. The EU record notes that Article 50(1) requires people to be told when they interact with an AI system, which reaches client-facing chatbots, while consent to AI use on a matter remains a national question. Germany’s BRAK guide reaches the same point from the other side: no professional duty under the BRAO or BORA, but transparency duties may still arise from contract law.

Sources