There Is No Privilege to Fall Back On

Staff will use AI. The useful question is which work may use a public tool, and which work must stay under the firm’s control. This briefing is a practical map for that line, as at August 2026. It is not professional or legal advice. The named codes and guidance are listed at the end.

1Why form-filling is the high-risk job

A research question can sometimes be written without names. A tax computation, an SSM filing or a KYC pack cannot. Those documents are the confidential record: identity, amounts, treatments, tax references. That is the material the ethics codes say must not leave the practice.

What most practices think they have done is “train a GPT” on house templates. What happens is retrieval. The knowledge base fetches the workpaper. The public model then reads the numbers. Paid ChatGPT, a Custom GPT, Copilot or an SME wrapper does not change that.

2A working rule for the firm

Draw three buckets and put every AI use into one of them. Then tell people which tools belong in which bucket.

1. Public AI. General learning and non-confidential questions. No client names, tax references, NRIC numbers or working-paper text.
2. Approved everyday tools. Firm-approved public AI for work that is not a confidential client file, inside clear rules. This is not a place for this year’s return or a KYC pack.
3. Private AI. Confidential client numbers: research, review and drafting on the file itself. This needs processing the office controls, a wall between engagements, and a person still signing what leaves. Secure Accounting is that path when the numbers themselves must stay inside the firm.

A ban on public AI is not a policy. People still need a way to get the work done. The rule above is the policy: public tools stay useful, and client numbers stay under the firm’s control.

3What the codes already require

The professional codes are not a software manual. They already describe the duties a first AI policy has to meet.

MIA’s Ethical Guidelines on Technology Usage (2023) tell members to keep information confidential across collection, use, transfer, storage, dissemination and lawful destruction, and to be conversant with the PDPA. There is no exception for AI tools. The MIA By-Laws absorbed IESBA’s technology-related revisions from 15 December 2024. Putting client names, NRIC numbers or tax references into a US chatbot is a transfer of confidential information to a third party. Under the PDPA it is also a cross-border transfer.

In Singapore, ISCA revised EP 100 on the same date, and ACRA made the matching changes to the Code for public accountants. The practising certificate asks whether the accountant has enough information and expertise to use and explain the technology, and whether using it creates self-interest or self-review threats. “The model produced it” is not an answer to those questions. The member remains the author. Wrong figures in a filing are still the member’s.

ICAEW’s tax-practice guidance (January 2026) states the same point in plain words. Putting client data into a publicly available AI tool is likely a confidentiality breach unless the client has consented. Input should be anonymised so the client cannot be identified. Once information is entered, control over that data is relinquished. That is UK professional guidance, not a Malaysian or Singapore statute. The codes in this region sit on the same IESBA technology revisions. The wording is useful because it names the disclosure.

In practice: do not put this client’s numbers into a public tool. Check anything that will be filed. The person who signs remains accountable.

4Data protection sits underneath the code

The Personal Data Protection (Amendment) Act 2024 raised the penalty for breaching the data-protection principles, and made data processors directly liable for the Security Principle. A prompt that contains client names, NRIC or tax references sent to a US model is a transfer under section 129.

KYC and beneficial-ownership packs are still confidential personal data. A public model is a third party. That is a disclosure question first, before it is an AML one. The member’s AML duties do not disappear because the step felt like administration.

5What to put in place

A first policy does not need to be long. It needs to be usable.

1. Write the three buckets down. Tell staff which tools sit in each one, and who to ask when the work is unclear.
2. Do not point a public chatbot at last year’s returns, house templates or a shared store over every client. That is how Client A’s numbers appear in Client B’s draft.
3. Require a person to check amounts, treatments and identities before anything is filed. Invented cases get the headlines. The filing is the everyday risk.
4. Keep a short record for higher-risk use. A client or an insurer who asks where their numbers were read needs an answer the firm can show. ICAEW already says consent is required before client data goes into a public tool.
5. Give confidential client numbers a private path: processing on premises, under the firm’s control.

The law-firm counterpart is Public AI on a Client File.