Blog · Accounting & AI · · 1 min read

Anonymize a tax return before ChatGPT

An accountant pasting a tax return or balance sheet into ChatGPT to check a line item has become common. The problem is that the client name, its SIREN, its amounts and its IBAN go with it, to a third-party service. The good practice is simple: anonymize the file before the AI, keep the key, and re-identify the result afterwards.

Original
Company : SARL Dupont & Fils SIREN : 552 100 554 Net income : 142 580 EUR Director : Paul Dupont IBAN : FR76 3000 4008 1200 0123
Sent to the AI
Company : [ORGANISATION_1] SIREN : [SIREN_1] Net income : [MONTANT_1] Director : [PERSONNE_1] IBAN : [IBAN_1]

What to mask in an accounting file

  • The client and directors' names.
  • The SIREN, SIRET and VAT number.
  • Identifying amounts (net result, capital, balances).
  • IBAN and bank détails.
  • Addresses, emails and phone numbers.

The method, without changing your tools

You pseudonymize the file, work with the AI on the masked version, then re-identify the result locally. The AI analyzes the structure and figures without ever knowing which client it is. You can also generalize amounts into ranges if you want to hide the exact values.

Safe-Doc pseudonymizes before the AI, handles PDF, Word and spreadsheets, keeps the layout and processes in the European Union then purges. Détails on the Safe-Doc for accounting page.

Test on a real accounting file, réversible pseudonymization, layout preserved.

Part of the guide : By profession ↗