AI for accountants and accounting firms: confidentiality and the three principles
Short answer
Accountants may use generative AI, including in the audit, as long as they meet their professional rules. The duty of confidentiality in their code of ethics still applies to client data, including towards a software vendor. The Dutch regulator AFM names three conditions: human involvement, traceable outcomes and controlled security. An AI environment under your own control makes that last one easiest to demonstrate.
AI does not change the duty of confidentiality
Confidentiality is one of the fundamental principles in the IESBA Code of Ethics and in the national codes based on it. What an accountant learns through an engagement is not shared with third parties without need. An AI service that receives annual accounts, ledgers or payroll records is such a third party.
So the question is not whether AI is allowed, but what happens to the data: how long it is kept, who can access it, whether it is used for training. See ChatGPT and the GDPR.
The three principles
| Principle | What it means in practice |
|---|---|
| Human involvement | The accountant assesses the outcome and remains responsible for the opinion |
| Traceability | It can be established which model, with which input, produced which outcome |
| Controlled security | Clarity on retention, encryption, updates and the vendor’s use of data |
The Dutch AFM does not rule out free AI tools, but asks that their use be controlled and documented. For audit client data, that is hard to demonstrate with a public service. See also AI hallucinations.
Where AI helps in accountancy
- reading contracts and minutes for items relevant to the audit;
- flagging anomalies in ledger entries for further testing;
- drafting management letters and disclosures;
- answering client questions about schemes from the firm’s own knowledge base;
- extracting invoices and receipts for bookkeeping.
Record per application who checks the outcome, in a firm-wide policy. See Writing an AI policy.
Client data under your own control
With werqly, AI runs on a node that belongs to your firm alone. Client data is processed there, not at an external AI vendor. The audit log records per user which model processed which source, supporting traceability, and in werqly AI Control you set sources and rights per team. Which questions to ask any AI vendor is covered in Assessing an AI vendor.
Frequently asked questions
Questions on this topic
May accountants use ChatGPT?
Regulators such as the Dutch AFM do not rule out free AI tools, but require human involvement, traceability and controlled security. The duty of confidentiality also applies. For client data, a public service is hard to justify on that basis.
Does confidentiality also apply towards an AI vendor?
Yes. An AI vendor that receives client data is a third party. Share only what is necessary, under appropriate agreements, or process the data in an environment that belongs to your firm alone.
What does traceability mean here?
That it can be established afterwards which model, with which input and settings, produced an outcome. That requires logging of use and a recorded model version.
Can AI carry out the audit itself?
No. AI can prepare work and flag anomalies, but professional judgement and responsibility remain with the accountant.
Sources
- IESBA — International Code of Ethics for Professional Accountants
- NBA — Three principles are the condition for AI use (November 2025, Dutch)
This article is general information, not legal advice. Rules and guidance change; check the sources listed or ask your lawyer or DPO if in doubt.