AI and Professional Secrecy: What Dutch Rules Allow

Dutch lawyers, notaries, accountants and doctors each have a statutory duty of confidentiality, and each of their professional bodies said something about AI between late 2024 and mid-2026. Put those texts side by side and they land on the same point.

AI and professional secrecy come down to one question: does what you enter into an AI system stay within the circle of people allowed to know it? With a public chatbot, the answer is no, and the professional bodies say so. With a business cloud licence, it depends on contractual safeguards you have to be able to demonstrate. With a model running inside your own firm's environment, the answer is yes, which is why the Dutch bar and the European bar federation both name that route explicitly.

What do the professional bodies say about AI and confidentiality?

The four Dutch professions with a statutory duty of confidentiality, and what their professional bodies have published on AI, side by side:

None of these bodies bans AI. They all ban the same thing: entering confidential data into a system when you don't know what happens to the input.

What does the Dutch bar tell lawyers?

A Dutch lawyer's duty of confidentiality sits in Article 11a of the Advocatenwet, the Dutch Act on Advocates, and extends to "employees and staff of the advocate, as well as other persons involved in the practice". The bar's own regulation, the Verordening op de advocatuur, has no provision on cloud or AI, only a general duty to organise the firm, case management and information management adequately.

The Dutch bar association (NOvA) filled that gap in December 2025 with its recommendations on AI in legal practice. Under confidentiality, they include: don't use confidential data in free tools, don't enter client data into public AI models, know in which country and with which subcontractors data is processed, and make sure "input and output stay within the firm's environment". Asked whether confidential client documents may go into a tool like ChatGPT, the NOvA answers: only with strong contractual safeguards, "or if you use AI tools that are hosted and managed within your firm".

The European bar federation CCBE says in its guide on the use of generative AI by lawyers of 2 October 2025 that a lawyer may breach professional obligations "already at the point of inputting the data". Alongside contractual terms, the safeguards it lists include "setting up AI systems to run locally or within a secured environment controlled by the law firm".

The NOvA texts aren't fully consistent on one point. The recommendations say to ask the client's permission for AI use in the file. The FAQ says there's no general rule requiring a lawyer to disclose every use of AI. A practical reading: put your AI policy in the engagement letter, and ask explicit permission when file documents go through an external system.

What does the KNB tell notaries?

Article 22 of the Wet op het notarisambt, the Dutch Notaries Act, imposes confidentiality on the notary and everyone working under their responsibility. The AI assessment framework of the Royal Dutch Association of Civil-law Notaries (KNB) offers a test that works for any profession: would you share this data with people or organisations outside your office? If not, the KNB says it's "not a good idea" to enter it into a model that processes it further.

On free off-the-shelf tools, the KNB is blunt: you run "a major privacy risk and may breach your duty of confidentiality", and it's "not advisable to use these systems (for now)". Its checklist also asks whether the AI system is located in Europe and managed from Europe. That second half is hard to answer with a confident yes for a US vendor with a European data centre.

What do the NBA and the AFM tell accountants?

A Dutch accountant's confidentiality duty sits in Article 16 of the VGBA, the professional conduct regulation: anyone who receives information they know or should suspect is confidential must keep it secret. In AI guide 2 of June 2026, the Dutch accountancy body NBA and the IT auditors' body NOREA write that client and personal data may not simply be entered into public AI models and must stay "in a secured, shielded environment".

The Dutch financial markets authority AFM links the data-leak risk of generative AI directly to accountants' confidentiality duty in its supervisory report 12 building blocks for controlled use of audit tooling (December 2025). It asks for a policy on minimisation, anonymisation, pseudonymisation and retention, both for processing in your own environment and at external parties. It isn't an absolute ban: in an NBA publication, the AFM says firms can use AI "as long as it's done in a controlled way".

What applies in healthcare?

Under Article 7:457 of the Dutch Civil Code, a care provider may not disclose information about a patient to others without consent, except to those directly involved in treatment. Article 88 of the Wet BIG adds a separate confidentiality duty for registered health professions. An AI vendor isn't a treating professional.

The Royal Dutch Medical Association (KNMG) says in its practice dilemma on ChatGPT (October 2024): never share personal data or commercially sensitive information with ChatGPT, because you can cause a data breach and breach medical confidentiality at the same time. It's not theoretical. In a warning from the Dutch data protection authority (AP) in August 2024, a GP practice employee had entered patients' medical data into an AI chatbot, against the practice's rules. The AP classes that as a data breach.

Is a business cloud licence enough?

A business cloud licence is sometimes enough. The NOvA and the CCBE both accept strong contractual safeguards, such as no storage of inputs and no training on your data. What those safeguards are worth in practice, the AP showed in July 2026. In a prior-consultation opinion on Microsoft 365 Copilot at the municipality of Haarlemmermeer, the AP concluded that the planned deployment "would infringe the GDPR", because there wasn't enough transparency about the service data Microsoft collects and how long it keeps it. The opinion concerned a municipality, not a law firm. The point is wider: a business licence is only a safeguard if you can show what happens to the data. In the accompanying guidance tool on generative AI, the AP itself writes that a system that shares personal data with the vendor and uses it for further training has "a greater impact on privacy" than a system the organisation has developed and manages itself.

Showing it is where local AI makes the difference. With an open-weight model on your own hardware, or in a private environment you manage yourself, there's no service data at a vendor, no subcontractors and no foreign jurisdiction. Why an EU region from a US vendor doesn't solve this is covered in EU data residency for AI.

What does the disciplinary tribunal ask?

A Dutch disciplinary case from July 2026 was mainly about something other than confidentiality. In ECLI:NL:TADRSHE:2026:93, dated 27 July 2026, a lawyer was reprimanded for court filings containing partly non-existent, AI-generated case law. The instructive part is what the dean of the bar asked as well: how had she safeguarded the confidentiality of client data in the AI tool she used? She answered that she hadn't entered any confidential data, and the disciplinary council found the dean's suspicion alone insufficient to establish a breach.

So the question will come. A firm needs to answer it with more than an after-the-fact statement: which system, which settings, what went in, who had access. With a local system, you record that in your own logs. On the privacy side, the AP also requires your own DPIA, even for a pilot, and says referring to another organisation's generic DPIA isn't enough. How to draw one up is in our DPIA for AI projects guide.

Which data can go where?

For professionals bound by secrecy, the choice comes down to the type of input, not the type of tool:

For sector-specific applications, see AI for lawyers and legal teams, AI for accountants and AI in healthcare. What your own environment takes, technically and financially, is in our guide to local AI for business.