Identifying data is replaced before any document reaches an artificial intelligence provider. The same person gets the same code in every document in the case file, so you keep the links between them. And when you need to see the real data, you recover it with your password.
You strike out the name on a contract and upload it. But the ID number is still in there, along with the address, the IBAN, opposing counsel’s bar registration number and the date of an accident that made the local papers.
You haven’t anonymised anything. You’ve removed one word.
And there’s a second, less obvious problem. If you anonymise document by document, each file gives the same person a different code. The claim says one thing, the defence another, the expert report another. The AI no longer understands that they are about the same case, which is exactly what you needed it to understand.
If Carlos Ruiz is TX320 in the claim, he is TX320 in the defence and in the expert report too.
Tools that anonymise file by file give him a different code in each one. That breaks the links between documents, and those links are exactly what a case file is made of.
The replacement doesn’t happen somewhere along the way: it happens before the document travels anywhere.
So there is never a moment when your clients’ data is out there unprotected.
The engine is rule-based. It recognises the identifying data in a Spanish legal document (the DNI national ID number, the IBAN, the bar registration number and the rest of the usual identifiers) and replaces it.
That is why we can say that no document reaches an AI provider before its data has been replaced. It isn’t a promise that depends on a setting: the path simply doesn’t exist.
If an AI did the detection, the document, with its data in the clear, would already have left your firm for that model to read it.
You choose a password when you start. With it, you recover the real data whenever you need it.
That password never leaves your browser. Our servers only receive a key derived from it, which is used during the request and then discarded. We don’t store it because we never receive it.
Without it, nobody can reverse the anonymisation, not even us. And that is exactly why, if you lose it, it cannot be reversed. We can’t help you recover it, because if we could, it would mean we could reverse it without you.
No, and be wary of anyone who promises that. ADOIA substantially reduces the personal data that leaves your firm, but regulatory compliance depends on far more than a tool. What we do make sure of is that identifying data doesn’t reach any AI provider.
No. In legal terms, anonymising means breaking the link with the person irreversibly: if that is achieved, the data is no longer personal data. Pseudonymisation replaces the identifiers while keeping the ability to undo it, and the data remains personal data under Article 4(5) of the GDPR. It is a security measure recognised by the Regulation itself, and it is usually the realistic option when you need to keep working with the document. We go into it in this article on ChatGPT in law firms.
No. It is a rule-based algorithm, with no model involved. That is deliberate: if an AI did the detection, the document, with its data in the clear, would already have left your firm for that model to read it.
The identifying data you usually find in a Spanish legal document: NIF, NIE, CIF, Social Security number, passport, bar registration number, IBAN, bank card, phone number, email address, vehicle registration, date of birth, dates, postal address, cadastral reference, case number, companies and natural persons. Identifiers and bank details are validated against their check digit, and variants of the same name are unified: “Juan Ruiz López”, “Juan Ruiz” and “Ruiz López, Juan” are the same person.
Yes. On top of those types, you can switch on a safety net that flags runs of capitalised words and alphanumeric codes. It is off by default on purpose, because it has low precision and gives way to any other detection, so it never duplicates what has already been recognised. It’s there for when you’d rather err on the side of caution.
You can mark it by hand before working with the document. And it’s worth knowing that no tool detects 100%: we tell you where our limits are before you start, not afterwards.
Not yet. Text only.
No, and that is deliberate. If we could recover it, we could reverse the anonymisation without you, and then the promise would not be true.
We show you how it works with your own documents, with the limits on the table, not under it.
15-day guided trial · No card · No commitment
