AI for financial advisors in Singapore: what to allow your team, and what to ban

Let advisers use AI for first drafts and internal notes. Keep client information out of public AI tools, and have a person read everything before it's used. Recommendations stay with the adviser.

MAS's Guidelines on Artificial Intelligence Risk Management, issued on 7 October 2026, give a basic list for lower-risk use that covers the same ground. The policy is your firm's to set, and the agency is where it gets followed.

As of November 2026.

MAS's examples of lower-risk use include drafting, proofreading or rephrasing emails to customers, and summarising documents or meeting notes for internal reference. Its footnote adds that people involved "should be reviewing and checking the outputs before using them."

• Client names with financial or medical details, policy numbers and bank details.
• NRIC numbers and copies. PDPC says organisations generally can't collect, use or disclose them without a legal requirement or a need to verify identity to a high degree of fidelity.
• A whole fact-find, whatever the reason for pasting it in.
• The recommendation itself. Which product suits a client is the adviser's call from that client's fact-find.
• Premiums, benefit figures and illustrations from memory or from a chat answer. Those come from the insurer's own system and documents.

Swap the name for "Client A" and remove anything that identifies the person. The draft usually works just as well.

MAS published the guidelines on 7 October 2026, and paragraph 1.8 has them take effect on 7 October 2027, with the life cycle and capability sections due by 7 October 2028. MAS's page lists financial advisers and life insurers among those they apply to, so they land on your firm first and reach you through its policy.

For lower-risk use, MAS allows basic governance if poor performance of the tool is unlikely to have a material adverse impact on the firm or its customers. That basic set should include:

• Someone clearly accountable for AI oversight.
• Permitted and prohibited uses, with MAS's own example being "prohibitions on inputting confidential, proprietary or client information into public AI tools".
• When a person has to review AI output before it's used.
• An approved list of tools, and a way to ask for a new one.
• Staff education and regular compliance checks.

MAS also names "use of shadow AI" as a human-factor risk. Its response to feedback describes that as public AI tools used by staff, and says the difficulty of finding them doesn't by itself justify leaving them out. Ask your advisers what they already use, because your firm will want to know.

Advice is treated differently. MAS lists provision of financial advisory services among the areas where controls on transparency, explainability and fairness may matter more, so AI that shapes a recommendation isn't the low-risk case.

• Section 24 of the PDPA requires "reasonable security arrangements" against unauthorised access, use and disclosure.
• Section 26 bars transfers outside Singapore except under requirements that give comparable protection, so ask where a tool processes its data.
• PDPC's Advisory Guidelines on Use of Personal Data in Generative AI, issued 20 July 2026, say organisations deploying these systems should educate users on the types of personal data that should go in, and encourage clear written policies. The guidelines are advisory.

Section 36 of the Financial Advisers Act says a licensed financial adviser must not make a recommendation on an investment product, to someone who may be expected to rely on it, without a reasonable basis. That basis has to come from consideration and investigation having regard to the client's investment objectives, financial situation and particular needs.

A chat answer doesn't do that work, so have your advisers do it from the fact-find. MoneySense tells clients the same thing from their side: a recommendation is only as good as the information behind it.

Put this in front of your firm's compliance team first, then share it with the agency. Fill in the brackets.

We use AI for first drafts and internal notes. Anything a client will see gets read by the adviser first.

Tools: use only these: [tools your firm has approved]. To ask for a new one, message [name].

Never put into a public AI tool: client names with financial or medical details, NRIC numbers, policy numbers, bank details, or a fact-find.

Never ask AI for: a product recommendation for a client, a premium, a benefit figure or an illustration.

Posts and ads: get [firm's approval process] before anything goes out, whoever drafted it.

Recording a client meeting: tell the client first, and use only [approved tool].

Questions go to [name]. We review this rule every [period].

• Has your firm approved it for client information?
• Where is the data processed and stored?
• Is what you type kept, and is it used to train the model?
• Can a client's data be deleted on request?
• Who can see it, at the vendor and at your firm?

Without clear answers to the first two, the tool is only for work with no client information in it. Get the answers in writing from your firm or the vendor.

• Assuming nobody on the team uses AI because nobody has mentioned it.
• Treating a confident answer as a correct one, especially on product terms or tax.
• Letting a note-taker's summary stand as the record when nobody has read it.
• Posting AI-written content before your firm has approved it.
• Asking AI to write a recommendation, then signing it.

Palby Sales Tracker's AI sits on the coaching side of your job. An AI chat can write an opener and two questions for a 1:1, next to a card with the adviser's last two weeks against their goal. In weekly check-ins, the writer can ask an AI coach about their week, and managers get a weekly summary.

The rules above still apply to whatever people type in. See how one-on-ones work.

• MAS, Guidelines on Artificial Intelligence Risk Management (7 October 2026), read 2 November 2026
• MAS, Response to Feedback Received on Guidelines on Artificial Intelligence Risk Management (7 October 2026), read 2 November 2026
• PDPC, Advisory Guidelines on Use of Personal Data in Generative AI (issued 20 July 2026), pdpc.gov.sg, read 2 November 2026
• PDPC, Advisory Guidelines on the PDPA for NRIC and other National Identification Numbers (31 August 2018), pdpc.gov.sg, read 2 November 2026
• Personal Data Protection Act 2012, Part 6, sections 24 and 26, read 2 November 2026
• Financial Advisers Act 2001, section 36, read 2 November 2026
• MoneySense, Financial advisory process, read 2 November 2026