Disclaimer: This article is for general information purposes only and does not constitute financial advice. Your personal circumstances are unique. Always seek advice from a FSPR-registered financial adviser before making financial decisions.
AI chatbots don't fail the way bad websites fail. A bad website looks bad. An AI answer about your KiwiSaver looks exactly like a good one — same confident tone, same tidy formatting — right up until you act on it. Here are the specific failure modes New Zealanders should know about, with real examples, and what happens (or rather, doesn't happen) when it goes wrong. This post is part of our AI vs financial adviser guide.
Failure Mode 1: Outdated KiwiSaver Rules and Rates
LLMs are trained on snapshots of the internet, so their "knowledge" freezes months or years before you ask the question. New Zealand's rules keep moving. The clearest recent example is the Budget 2025 KiwiSaver overhaul:
What Changed (and What a Stale AI May Still Tell You)
Sources: Budget 2025 — KiwiSaver and Inland Revenue — KiwiSaver changes, as at July 2026.
If you planned your contributions around the old $521.43 figure because a chatbot quoted it, you'd be working from numbers that stopped being true in mid-2025. Nothing in the AI's answer signals this — stale answers read identically to current ones.
Failure Mode 2: Invented Fund Figures and NZ "Facts"
Hallucination — generating plausible-sounding content that isn't true — is a well-documented property of LLMs, and NZ finance is exactly where it bites, because the training data is dominated by overseas content. Patterns to watch:
Precise-looking numbers with no source
Ask for "the average KiwiSaver fee" or "Fund X's five-year return" and you may get a specific percentage that appears nowhere in any provider's documents. The precision is the trap: a made-up "1.27% p.a." feels more credible than an honest "it varies". Real fund data lives in product disclosure statements and tools like Sorted's fund comparison — not in a model's memory.
Confused institutions and mixed-up roles
AI answers routinely blur the FMA (the regulator) with the FSPR (the Companies Office register), or cite rules from Australia's regime as if they applied here. If an answer says an adviser is "registered with the FMA", that's a tell — registration happens on the FSPR, licensing with the FMA.
Imported US and Australian assumptions
Advice frameworks built around 401(k)s, Roth IRAs, US capital gains tax or Australian superannuation leak into NZ answers constantly. New Zealand's tax treatment of investments (PIEs, PIRs, the FIF rules) is genuinely different — a strategy that's textbook-sensible in the US can be tax-inefficient or simply meaningless here.
Failure Mode 3: No Duties, No Suitability, No Push-Back
A licensed adviser operates under the Code of Professional Conduct for Financial Advice Services and the FMC Act's adviser duties: they must understand your circumstances, ensure you understand the advice, exercise care, diligence and skill, and give priority to your interests when a conflict arises. Those duties shape behaviour in ways you can feel in the room — a good adviser interrogates your assumptions before endorsing your plan.
A chatbot does the opposite. It answers the question as framed. Tell it you want to put your house deposit into a single volatile asset and it will often help you plan the purchase rather than challenge the premise. It has no duty of suitability because it has no duties at all.
Failure Mode 4: When It Goes Wrong, There Is No Recourse
This is the risk that compounds all the others. If a licensed adviser gives you unsuitable advice, you have a ladder of remedies:
- Complain to the adviser's firm (their disclosure statement must set out the process).
- Escalate free of charge to their approved dispute resolution scheme — such as Financial Services Complaints Ltd (FSCL) or the Insurance & Financial Services Ombudsman (IFSO) Scheme — which every FSPR-registered adviser serving retail clients must belong to.
- Report conduct concerns to the FMA, which can investigate and act against licensees.
If ChatGPT's answer costs you money, that entire ladder is missing. The provider is an offshore technology company whose terms disclaim professional advice; there is no scheme, no licensee, and realistically no remedy. You absorbed 100% of the risk the moment you acted on the output.
Failure Mode 5: Your Data Goes Somewhere a Chatbot's Terms Decide
To get a "personalised" answer out of a chatbot, people paste in the personal details: income, debts, KiwiSaver balance, sometimes whole bank statements. Depending on the provider and your settings, that information may be retained and used to improve the service, and it sits with an offshore company under its own terms — not with a licensed NZ adviser whose conduct obligations, complaints process and record-keeping the FMA can examine. Before typing financial details into any AI tool, strip out identifying information (IRD numbers, account numbers, full names) and assume anything you paste could persist. A licensed adviser asking for the same information is doing so under a regulated relationship; a chatbot is not.
Using AI Without Getting Burned
None of this means "never use AI for money questions". It means: use it for education and preparation, verify every factual claim against primary sources — our step-by-step fact-checking checklist shows how — and keep actual decisions with a professional who owes you legal duties. If you're weighing up a licensed digital alternative, see our robo-advice NZ guide; if you're ready for a human, start with how to choose a financial adviser in NZ and verify anyone you shortlist on the FSPR.
Get Answers Someone Is Accountable For
We connect you with FSPR-registered financial advisers across New Zealand — professionals with disclosure duties, conduct standards and a dispute resolution scheme behind them.