Using AI at a law firm without leaking client data
This is a practical guide, not legal advice. If your organisation has a data protection officer or a privacy lead, their rules come first.
Why it weighs heavier for lawyers
Lawyers work with information that falls under professional secrecy. As soon as a pleading, an email thread or a contract goes into a public AI service, that information is in the hands of a third party. A name, an amount and a short description of a case together can be enough to identify a client.
What this means in practice
- Choose the account. A business plan with a data processing agreement carries a different risk than a free consumer account. Check what the provider does with your input.
- Paste the shape, not the data. To rewrite a paragraph, structure a defence or summarise a long email thread, the model does not need the real names and amounts.
- Add your own terms. Put client names, opposing parties, case numbers and addresses on the own list of Scrub for AI, and they are replaced everywhere in the text.
- Read before you send. A name in a free sentence can stay behind. Check the text before it goes out.
What Scrub for AI does here
Scrub for AI replaces recognisable data in a text with placeholders such as [EMAIL_1], [IBAN_1] and [PHONE_1], on the user's own computer and before the text reaches an AI chat. One click puts the real values back in the answer. It finds patterns (email addresses, phone numbers, IBANs, card numbers, Dutch BSN and VAT numbers, dates of birth after a keyword, secrets in logs) and the terms you add yourself, such as client names or case numbers.
It lowers the risk and is not a guarantee. Names and free descriptions are only found when they are on your own list, and you always read the result before you paste it.