AI and privacy
How a law or notary firm uses PrivaSearch and its own AI assistant to search contracts, litigation files and deeds, with confidentiality in mind.
A law or notary firm uses PrivaSearch to search contracts, litigation files and deeds with an AI assistant, without giving the names and addresses of clients and opposing parties to the provider. Documents are put in a project per case, pseudonymised, checked by a colleague and then made searchable with source references.
In this article you see what that looks like in practice, based on made-up situations. We start with the rules you have to deal with.
For lawyers, the duty of confidentiality is laid down in Article 11a of the Dutch Advocates Act (Advocatenwet), for notaries in Article 22 of the Dutch Notaries Act (Wet op het notarisambt). Both duties apply to everything you learn in your professional capacity, and also to the people who work under your responsibility.
The Netherlands Bar (NOvA) published Recommendations on AI in the legal profession in December 2025. Among other things, they say: do not use confidential data in free tools, do not enter confidential or client data into public AI models, and ask the client for consent to the use of AI in the file. The NOvA also asks that quotations, case law and facts are always checked manually.
The Royal Dutch Association of Civil-law Notaries (KNB) published the AI scales (AI-weegschaal), an assessment framework with a practical test: would you share this data outside your firm? If not, you do not enter it into a model that processes the data further. For the time being, the KNB advises against free, generally available tools.
What generally happens when someone puts a client file into a chatbot is set out on our page about AI chatbots and the GDPR. Position as of October 2026. This is an explanation, not legal advice.
PrivaSearch is software that lets your team search its own client files with Claude, ChatGPT or Mistral Vibe, after names, BSNs (Dutch citizen service numbers) and addresses have been replaced by pseudonyms and a colleague has checked that.
The pseudonyms are consistent. Where the same opposing party appears in the writ of summons, the exhibits and the correspondence, the assistant reads the same pseudonym everywhere, for example [PERSOON_B4A1]. That way it can make connections between documents without knowing who it is.
A lawyer takes over a tenancy case from a colleague. The file contains the lease, two addenda, a report on overdue maintenance and years of emails. The client's question: when did the landlord first respond to the complaint about the leak? In PrivaSearch, the lawyer asks about the first response to the complaint. The answer comes with the passages and the documents they are in. The lawyer rereads them before building a timeline on them.
A client wants to know which of its supplier contracts contain a penalty clause or a unilateral right to amend. Via PrivaSearch, an assistant can search all released contracts in the project and list the provisions with source references. Contract amounts and terms remain readable; names of contact persons have been replaced. The lawyer assesses the provisions personally and writes the advice.
A notary's office is preparing a deed of transfer and wants to check which easements and chain clauses appear in earlier deeds of the plot. Those deeds are full of names, addresses and dates of birth of previous owners. After pseudonymisation and review, an employee can ask the assistant which clauses are in the earlier deeds, with the page attached. The notary checks the text in the deed itself.
When an estate is settled, wills, bank statements and correspondence with heirs come together. Searching for "which accounts are mentioned" or "what arrangements are there about the house" saves time. Note: a will can contain information that is not personal data in the technical sense, but is strictly confidential. Only put in a project what is necessary.
Pseudonymisation protects the identifiability of persons. Professional secrecy goes further: it applies to everything you learn in your professional capacity, even without a name attached. A passage in which names have been replaced can still describe a case that someone in the sector recognises.
| Question | What PrivaSearch does | What you decide |
|---|---|---|
| Are names, addresses and BSNs shielded from the AI? | Replaces them with pseudonyms; a colleague checks | Whether the review is complete |
| Does the AI see the original file? | No, only released passages, never originals or file names | Which documents belong in a project |
| Are amounts, dates and descriptions shielded? | No, they remain readable | Whether a document is too recognisable to share |
| Is the AI's answer correct? | Gives the source per passage | Rereads and checks |
| May AI be used in this client file? | Does not make that decision | With the client and in line with your firm's policy |
Pseudonymised data remains personal data for your firm. We explain the difference in Pseudonymise or anonymise?.
Connect only a business account. What PrivaSearch returns to the assistant reaches the provider of that assistant, with pseudonyms instead of names. So connect a business account with a data processing agreement. What you arrange in it is set out in our article on the data processing agreement with an AI provider.
Do not type names into your assistant. What you type into the assistant yourself goes directly to the provider, bypassing the pseudonymisation.
Work per case. Permissions apply per project. A colleague who is not in the project sees nothing of the case, and administrators do not get automatic access. That way you also keep a Chinese wall within the firm in place. Give every project an end date: after that, access is closed and data is cleared away, unless a legal hold applies.
Draw up a firm policy. The NOvA recommends a firm-wide AI policy and asks firms to inform clients about it. For this processing, PrivaSearch supplies an AI policy and a DPIA starter document, derived from the set-up of your environment; you tailor them to your firm yourself. What a DPIA for AI tools looks like is explained in The DPIA for AI tools.
The steps are also on the page How PrivaSearch works. If you want to see it with a client file of your own, book a demo.
Start small and with a kind of question that comes up often at the firm. For a law firm, that is often finding facts and arrangements in an extensive client file; for a notary's office, rereading clauses in earlier deeds. Choose one or two closed client files, put them in a project and have a colleague do the review. Then you see how long the review takes per document and what detection misses.
Then record how the firm works with it: which kinds of documents may go into a project, who does the review, which assistant is connected and how you inform clients. That belongs in the firm's AI policy, which the NOvA also recommends.
Make sure that everyone who searches has completed the short AI literacy training. Without a valid training, the workspace does not open and the connection with the assistant is refused as well. Why that training exists is set out in our article on AI literacy.
Evaluate after a few weeks. Does searching with sources save time, and were the answers right when you checked them? Only then expand to ongoing cases.
Yes, but the NOvA recommends not using confidential data in free tools, not entering client data into public AI models and asking the client for consent to the use of AI in the file. The lawyer remains ultimately responsible and checks quotations, case law and facts personally. A way of working in which personal data is shielded first and a human checks fits with that.
The KNB published the AI scales (AI-weegschaal), an assessment framework for notaries. A practical test from it: would you share this data outside your firm? If not, do not enter it into a model that processes it further. For the time being, the KNB advises against free, generally available tools.
Not if you search via PrivaSearch. Names, addresses, BSNs, IBANs, number plates and dates of birth are replaced by pseudonyms within PrivaSearch, and a colleague checks that in the review. The link between pseudonym and person is kept in a separate vault; revealing is only possible in PrivaSearch itself, with a permission check and a log entry.
Not automatically. Pseudonymisation shields persons, but professional secrecy applies to everything you learn in your professional capacity. A case can remain recognisable through amounts, dates or descriptions. So decide per client file which documents belong in a project and whether the use of AI fits the arrangements with the client.
An AI assistant can make mistakes, even when it works with good sources. PrivaSearch gives the source with every passage, so that you can check whether a statement is in the client file. For case law and statutory texts outside the client file, you check the original source yourself, as the NOvA also recommends.
Yes. You connect a SharePoint library read-only, so that nothing changes in your source systems. Every request checks live whether the user may still see the document in SharePoint. Permissions revoked in SharePoint means that the document is no longer visible in PrivaSearch either.
For each Claude plan: what happens to your input, how long it is kept, whether an agreement is in place and where your data is stored. October 2026.
What you need to arrange for each type of data before using an AI chatbot: from names and email addresses to health, criminal records and BSNs.
Where an uploaded file ends up, how long it stays, who can access it and what deletion means. With four approaches for files containing personal data.
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 →