Rules
What confidentiality means when you put a client file in an AI chatbot, and what the NOvA, KNB, NBA and CCBE have published about this practice.
Not without further consideration. Putting a client file in a chatbot gives its contents to a third party: the provider. A free or personal account has no contract suited to your duty of confidentiality, and the professional bodies advise against it. A business service, confidentiality arrangements and your own assessment allow more, but responsibility remains yours.
This article covers professional rules. What the GDPR says about the same action is on our page about AI chatbots and the GDPR and in the article about client data in an AI chatbot. Both apply: processing may be justifiable under the GDPR yet breach your confidentiality duty, and vice versa.
The GDPR protects personal data. Professional secrecy protects the trust that lets someone speak freely with their lawyer, notary or accountant. This creates three differences that matter for AI.
Pseudonymisation therefore helps with the GDPR and people's identifiability, but does not solve every confidentiality issue. A passage with names replaced may still describe a case someone in the sector recognises. Read more about that boundary in our article on pseudonymisation or anonymisation.
Under Article 11a of the Dutch Advocates Act, unless the law provides otherwise, lawyers must keep confidential everything learned through practising their profession. The same duty applies to employees, staff and other people involved in that practice. The second paragraph states that the duty continues after professional practice or employment ends. Article 10a of the same Act lists confidentiality as a core value, alongside independence, acting in the client's interests, expertise and integrity.
Article 22 of the Dutch Notaries Act follows almost the same structure. Unless the law provides otherwise, notaries must keep confidential everything learned through their work. This also applies to people working under their responsibility and continues after office or employment ends.
For accountants, the duty is in the regulation governing accountants' conduct and professional practice, the NBA's VGBA. Confidentiality is one of its five fundamental principles (Article 2). Article 16 requires accountants to keep confidential data or information whose confidential nature they know or should reasonably suspect, with a limited number of exceptions, such as a legal obligation or the client's written permission for a specific purpose. Article 18 prohibits using confidential data for personal gain or a third party's gain. Article 19 requires measures to ensure that people working under your responsibility, or whom you ask for support, also maintain confidentiality.
The text we consulted is the version the NBA publishes as effective from 1 January 2020. Check with the NBA whether anything has changed since then.
In many cases, lawyers and notaries can also invoke their confidentiality against courts and investigating authorities. The position is different for accountants. This article does not explore that further: it is a separate subject, and a specific case requires a lawyer. For AI, the main issue is where information ends up and who can then access it.
Below is what the bodies have published, with dates. The position is as of 13 October 2026. These are living documents: the NOvA itself says new insights may lead to revisions.
| Profession | Basis for confidentiality | What the professional body says about AI |
|---|---|---|
| Lawyer | Article 11a Dutch Advocates Act; core values in Article 10a | NOvA, Recommendations on AI in legal practice (December 2025): no confidential data in free tools, no client data in public AI models, client permission for AI use on their file, a firm-wide AI policy |
| Notary | Article 22 Dutch Notaries Act | KNB, AI scales (November 2025) and part 2 of its assessment framework (December 2025): general free tools discouraged, establish what the provider does with input, do not enter what you would not share outside the firm |
| Accountant | VGBA, Articles 2 and 16 to 19 | NBA and NOREA, Guide 2: AI applied (June 2026): do not simply enter client and personal data in public AI models; data stays in a secured, restricted environment |
| Lawyer (European) | CCBE charter and model code of conduct | CCBE, guide on generative AI (2 October 2025): do not enter client data unless appropriate safeguards exist; transparency towards clients who would reasonably object |
The Netherlands Bar announced its Recommendations on AI in legal practice in a news item dated 16 December 2025. They are organised by core value. Under confidentiality, they include:
The other core values state that the lawyer retains ultimate responsibility, always checks quotations, case law and facts manually, draws up a firm-wide AI policy and informs clients about it. The document is called “recommendations” and provides guidance. The law, core values and supervision by the local bar president also continue to apply. We cannot predict how a local bar president or disciplinary tribunal would weigh the recommendations in a specific case.
On 20 November 2025, the Royal Dutch Association of Civil-law Notaries published its AI scales, an assessment framework showing the factors a notary considers. Part 2 followed on 18 December 2025, with specific questions for the notary and software supplier.
Part 2 of the assessment framework includes a chapter on protecting data and maintaining confidentiality. Two passages are useful to remember in practice. The first is a test: would you share this data with people or organisations outside your firm? If not, the KNB says it is not a good idea to enter it into a model that processes it further, because you lose control over the input and risk breaching your confidentiality duty. The second concerns free, generally available tools: you face a major privacy risk and may breach confidentiality, and the KNB currently advises against using them.
The NBA published the AI in Control guide in November 2024 and, with NOREA, Guide 2: AI applied in June 2026. The second guide's first precondition is that ethical and professional frameworks continue to apply in full, particularly the VGBA's fundamental principles. On confidentiality, it says client and personal data must not simply be entered into public AI models, giving free versions of well-known chatbots as examples, and that this data must stay in a secured, restricted environment. The professional retains full responsibility; according to the guide, AI output alone is never sufficient audit evidence.
In a November 2025 interview on the NBA website, two supervisors from the Dutch Authority for the Financial Markets (AFM) name three conditions for accounting firms' use of AI: people remain involved, outputs are traceable and security is controlled. They offer questions any firm can ask: how long does the supplier keep client data, is it properly encrypted, and does the supplier do anything else with it?
On 2 October 2025, the Council of Bars and Law Societies of Europe (CCBE) published a guide on the use of generative AI by lawyers. The core rule: do not enter personal, confidential or other client data unless appropriate safeguards exist. Examples in the guide include a contractual duty for the provider to treat data confidentially or retain nothing, a data processing agreement limiting use to the firm's purposes, and technical measures such as a system running locally or in a dedicated secured environment.
On the client, the CCBE is less categorical than the NOvA. If an informed client could reasonably be expected to object or set conditions, the lawyer must be transparent. The NOvA recommends asking permission.
The publications differ in tone but point the same way. In daily practice:
If something does go wrong, for example a colleague puts a file into a personal account, assess whether it is a data breach. This is separate from whether you breached professional rules; you must answer both questions.
PrivaSearch sits between your documents and the AI model. Personal data is detected within PrivaSearch and replaced with consistent pseudonyms. A colleague sees the original alongside the pseudonymised text, corrects what the model missed and approves it. Only then is a document searchable. The steps are on How PrivaSearch works.
An AI assistant searching through PrivaSearch receives only released, pseudonymised passages with a source reference. No original files, filenames or links. That connection works with Claude, ChatGPT Business and Mistral Vibe. You can also search entirely within PrivaSearch, without a connected assistant. Access is per project: only project members see its contents, and actions are logged.
This fits what the professional bodies require: knowing which data goes where, sharing less than the whole file, and human checking. It does not replace your own judgement. Pseudonymised data remains personal data for your organisation. A passage without names may still contain confidential case information, so whether your client agrees and whether it fits your confidentiality duty remain your questions. Anything you type yourself into a connected assistant goes directly to its provider.
Our measures for each part of the GDPR and AI Act are described on the GDPR and AI Act page. To see it with a fictional file from your own practice, book a demo.
The NOvA advises against a free or personal account: no confidential data in free tools and no client data in public AI models. A business service allows more if you know what the provider does with data, have checked the contractual terms and involved the client. The NOvA recommends asking permission to use AI on the file.
That depends on the tool and agreements. The KNB warns that free, general tools may breach your duty because you do not know what happens to input. The CCBE requires appropriate safeguards, such as a contractual duty of confidentiality and a data processing agreement. Assess whether your situation meets those requirements yourself, with your professional body or a lawyer if needed.
The NOvA recommends asking permission to use AI on the file and informing clients about the firm's AI policy. The CCBE requires transparency when an informed client would reasonably object or set conditions. We did not find such a recommendation for notaries and accountants in the documents we read, but transparency in your engagement letter prevents later disagreement.
The document is called recommendations and provides guidance for applying the core values. The confidentiality duty under the Dutch Advocates Act, the GDPR and the AI Act are binding. How a local bar president or disciplinary tribunal would weigh the recommendations in a specific case cannot be predicted. Treat them as the standard against which your conduct may later be assessed.
It is a precondition, not the endpoint. You receive a data processing agreement and your input is not used for training by default. The full file still goes to the provider, nobody checks its contents beforehand and the client does not automatically know about it. The professional bodies additionally require you to know data flows, share only what is necessary and check outputs.
Partly. It removes names, addresses and numbers from the text an AI model sees, reducing the risk for the people concerned. Professional secrecy goes beyond personal data: even a case without names may be recognisable or confidential. Pseudonymisation is therefore a measure alongside permission, contractual agreements and your own assessment, not a replacement for them.
When a decision involving AI falls under the ban on automated decision-making, which exceptions apply, and how a human genuinely takes part.
Privacy by design and by default for AI tools: what to settle when choosing a tool, which settings to switch off and what to ask of the supplier.
What the AIIA of the Dutch Ministry of Infrastructure is, who must use it, how it is structured and how it relates to the IAMA and the DPIA.
We show it with fictitious client files from your sector: upload, detection, review, search and the connection with your AI assistant. Afterwards you'll know exactly what your team can use safely.
Book a demo →