Review

When an AI insurance tool owes you a duty, and when it owes you nothing

Source: FMA, Thematic review: Artificial intelligence in financial advice

If an AI tool tells you which policy to buy, New Zealand law already treats that as regulated advice, and the firm behind it must be licensed and put your interests ahead of its own. If it only tells you what a policy says, none of that applies. The FMA has now started asking firms how they handle that line.

By Henry Smith · Regulation · 2026-08-07

What this means for you If you already hold life, health, trauma or income cover, this review changes nothing about that policy and adds no new protection to it. What it puts on the record is that the existing law applies to an AI tool exactly as it applies to a person: a tool that ranks named policies and badges one as right for you is giving regulated advice, so the firm behind it must hold a Financial Advice Provider licence from the regulator and put your interests ahead of its own. A tool that only reads out what a policy says carries none of those duties, and the FMA has published no guidance on where each AI-assisted step falls.

Written by Henry Smith · Reviewed by Smiths Insurance & KiwiSaver, licensed Financial Advice Provider (FSP712931)

Ask an AI tool which insurance policy to buy and, if it answers, New Zealand law treats that answer as regulated advice. The business behind it has to hold a licence, and whoever gives the advice has to put your interests ahead of their own. Ask the same tool what your policy's exclusions say and it owes you none of that. On screen, the two answers can look identical.

Nobody has mapped which AI-assisted steps land on which side of that line. On 6 August 2026 the Financial Markets Authority, the government body that licenses and polices financial firms here, started asking the industry how it is coping. One of the answers firms can tick on its questionnaire is that "AI outputs are used directly with customer without adviser review" (FMA, Financial advice providers and advisers survey).

What does this mean for me?

If you already hold life, health, trauma or income cover, nothing here changes that policy or adds anything to it. What it changes is how much weight to put on a screen, because the protections in the law attach to advice, not to information.

One question sorts it, and it works on any tool, ours included: is it telling you what a policy says, or telling you what to buy?

If it is telling you what to buy, the business behind it needs a Financial Advice Provider licence, and the law attaches duties to what it tells you: competence, care, and your interests first. If it is only telling you what a policy says, the decision and everything that follows are yours alone.

Where does the law draw the line?

Two parts of the Financial Markets Conduct Act 2013, the main law covering financial products and advice here, do most of the work, and neither mentions technology.

Section 431C says you are giving financial advice if you recommend, or give an opinion on, buying or selling a particular financial product. Telling someone not to buy, or not to sell, counts the same way. Insurance sits squarely inside this, because section 6 treats an insurance contract as a "financial advice product", so life, health, trauma and income cover are all in scope. Once advice is regulated, real duties attach: sections 431I to 431P require the person advising you to be competent, to put your interests ahead of their own, and to work with care, diligence and skill (Financial Markets Conduct Act 2013, sections 6, 431C and 431I to 431P and Schedule 5 clause 7).

Schedule 5, clause 7 then lists what is not financial advice. Two entries carry most of the weight. You are not giving advice merely by:

"providing factual information (for example, information about the cost or terms and conditions of a financial advice product, or about the procedure for acquiring or disposing of a financial advice product)"
"making a recommendation or giving an opinion about a kind of financial advice product in general rather than a particular financial advice product" (same source)

Two further entries, both in the table below, cover pointing someone to an adviser and passing on advice someone else gave.

We run an insurance comparison site, so this is our own problem to get right rather than a thought experiment. This is how we read the Act against what an AI-assisted comparison actually does.

What the tool does	The part of the law	Our reading
Shows a named insurer's published price, benefit limits or exclusions	Sch 5 cl 7(a), factual information	Information, not advice
Explains in general terms how trauma cover works	Sch 5 cl 7(c), a type of product in general	Information, not advice
Filters a product list using details you typed in	Sch 5 cl 7(a), if all it hands back is the filtered facts	Information, until it says the result suits you
Ranks named policies and badges one as right for you	s 431C(1)(a), an opinion on buying a particular product	Advice
A chatbot answers "should I switch from my insurer to this one"	s 431C(1)(a), buying or selling a particular product	Advice
AI drafts a recommendation and an adviser reviews and signs it	s 431C(1)(a), with ss 431I to 431P applying to the adviser and the firm	Advice from the licensed firm, whatever wrote the draft
Tells you to go and speak to an adviser	Sch 5 cl 7(d), recommending you get advice	Not advice
Repeats an adviser's recommendation without claiming it as its own	Sch 5 cl 7(e)	Not advice

Sourced to sections 6 and 431C and Schedule 5 clause 7 of the same Act. How the Act applies turns on the facts of each case, and this table is our reading, not a legal opinion and not the FMA's position.

The part that is easy to miss is that the trigger is the recommendation, not the robot. Plain software with no AI in it that ranks policies and badges a winner is giving advice. An AI chatbot that reads out a published exclusion is not. Clause 7 cares about what the output says, not about what produced it, which is why a tool can cross the line halfway through a sentence with nobody having decided that it should.

Why has nobody drawn the map?

Because the one body that could has not tried yet. The FMA's thematic review is a fact-finding exercise across a whole industry, not an investigation into one firm. Four surveys, going to advice firms, technology suppliers, lawyers and industry bodies, are open until 4 September 2026. The firm survey asks about "governance, oversight, suitability, record-keeping and consumer outcomes": who is in charge of the AI, who checks its work, and whether customers end up better or worse off (FMA, Thematic review: Artificial intelligence in financial advice, 6 August 2026).

Nobody is in trouble. The FMA calls this "an exploratory exercise rather than a formal review or enforcement initiative" (FMA, Engagement overview, August 2026). The same document sets out the regulator's starting position on the law: the existing framework, including the Act and the Code of Professional Conduct for Financial Advice Services 2025 that every adviser has to follow, "is technology-neutral". It also records a workshop with University of Auckland researchers on 15 April 2026, where more than 40 industry people raised "concerns about AI outputs blurring the distinction between the provision of factual information and giving regulated financial advice".

So the regulator has named the problem without solving it. In the same firm survey, question 34 asks which parts of the law leave firms most unsure about AI, and one of the sixteen options listed is the "distinction between regulated financial advice and other forms of guidance or information". Question 5 splits "research and product comparison" from "suitability assessment/ recommendations", the same divide, one step earlier. Nothing in the review is guidance.

Until that changes, the protection you get from a tool is decided by what its output says, not by how capable it looks or how confident it sounds. A tool that hands you facts leaves the decision, and what follows from it, with you. A tool that tells you which policy to take on is answerable for that, and if the business behind it holds no licence to give advice, it should not be making that call at all.

QuoteHub is operated by Craig Smith Business Services Limited, trading as Smiths Insurance and KiwiSaver, a licensed Financial Advice Provider, FSP712931. We read the public documents for this piece. We were not briefed, and we have not spoken to the FMA about the review.

What could we not check?

The FMA has published no guidance on where AI outputs sit against section 431C, and this review is expressly not an attempt to write any. The engagement overview says the FMA "may" publish a high-level summary of what it observes, talk to the industry further, or use what it learns to shape future guidance. Nothing is committed and no date is set.

The survey questions we quote come from PDF copies the FMA publishes "for reference purposes only". The FMA says the version in its survey portal is the definitive one, so the wording may differ. We could not read the responses, because they are not public, and the FMA says reports drawn from the questionnaire "will generally be aggregated and anonymised", meaning grouped together with no firm named.

None of this is legal advice. Whether a particular output counts as regulated financial advice depends on the whole of the interaction, and on further exclusions in Schedule 5 clauses 8 to 18 that this piece does not work through.

What this means for your cover

What a policy pays, how to size it, and how a rule change reaches an existing policy. Life insurance in New Zealand

Sources

Every source below was read and checked on 21 August 2026.

All QuoteHub insurance news

Talk to a licensed adviser about what this means for you · free, no obligation.