ASIC's 2027 AI enforcement focus: checks for NZ finance firms

By DataForge team20 November 20265 min read

ASIC will target AI and chatbot failures that harm consumers in 2027. NZ regulators already expect the same accountability, so firms should test their controls now.

What ASIC announced

On 18 November 2026, ASIC chair Sarah Court set out the regulator's nine enforcement priorities for 2027 at its Annual Forum in Sydney (Capital Brief). Priority four covers consumer harm from poorly implemented technology. ASIC's media release puts AI, automated decision-making and chatbots under that heading (ASIC 26-286MR). ASIC will look for serious failures where systems give customers misleading information or make inappropriate decisions. It'll also look at systems that stop people reaching effective support.

In her speech, Court said ASIC's attention will go to systems that are poorly designed or inadequately supervised. She was clear that AI tools "do not alter a firm's legal obligations" (ASIC speech). The speech names four situations where customers need access to a human reviewer: making a complaint, responding to a scam, pursuing an insurance claim, and asking for financial hardship help. No AI-specific penalty or case was announced alongside the priority (techpartner.news).

The AI item sits on a list topped by cost-of-living misconduct at banks, insurers and super trustees, such as discounts that weren't passed on and poor claims handling. The list also includes accountability for the Shield and First Guardian fund collapses (ASIC 26-286MR). Several of those areas overlap with AI. A claims triage model or a hardship chatbot that fails customers could breach both the technology priority and the cost-of-living priority at once.

Why an Australian enforcement list matters in New Zealand

ASIC's reach covers NZ firms that hold an Australian financial services or credit licence. It also affects NZ software and service providers whose Australian clients will now ask harder questions about how their tools behave. NZ's largest banks are Australian-owned, so we expect group-level AI controls built to satisfy ASIC to flow through to NZ operations.

ASIC's position also lines up closely with what NZ regulators have said in the past few weeks. On the same day as the ASIC forum, the Commerce Commission released guidance on how the Fair Trading Act and Commerce Act apply to AI. The Commission says a business remains responsible for outcomes whether a person, an algorithm or an AI system made the decision (Commerce Commission). The Commission also asks boards to treat AI as a governance issue and to know where AI influences decisions.

For financial services, the RBNZ and FMA said on 15 October 2026 that boards and senior management should know where AI is used, understand the risks AI creates, and be clear on who is accountable (FMA). The FMA's thematic review of AI in financial advice, launched on 6 August 2026, covers generative AI, machine learning, decision-support tools and agentic systems. Its surveys closed on 18 September and no findings have been published yet (FMA thematic review).

None of the NZ regulators has published an enforcement priority as specific as ASIC's. Even so, the direction on both sides of the Tasman is the same: existing conduct law applies to automated decisions, and regulators expect a named person to answer for each one. ASIC has now said publicly which failures it'll pursue first. We think NZ firms should treat that list as a reasonable preview of what the FMA and Commerce Commission will look at once their own reviews finish.

Where AI deployments tend to fall short

ASIC's wording points at the gap we most often see in customer-facing automation. Many teams test their chatbots and decision models for accuracy on common requests. Far fewer test what happens when a customer is in one of the four situations Court named. A chatbot that's built to deflect contact will look efficient on a dashboard while trapping someone who's trying to report a scam. A claims model can be accurate on average and still decline an unusual but valid claim with no route to a person.

Supervision is the other weak spot. ASIC singled out systems that are inadequately supervised, and that description fits deployments where nobody reviews outputs after go-live, where model or prompt changes skip change control, or where the vendor updates the underlying model without telling the customer. The RBNZ and FMA's reference to third-party risk applies directly here, because much customer-facing AI runs on vendor platforms the firm doesn't control (FMA).

Record-keeping is the third gap, and it carries its own risk because failure to meet record-keeping obligations is a separate item on ASIC's 2027 list (ASIC 26-286MR). If a firm can't reconstruct why a system declined a customer, the firm will struggle to show the decision complied with its obligations.

Practical next steps

These steps suit NZ firms in financial services, insurance and lending, and the software and service providers that support them. Most of them also apply to any business using AI with customers, given the Commerce Commission's guidance.

  • Map every customer-facing AI touchpoint. Include chatbots, triage models, credit or claims scoring and AI features built into vendor platforms. Record the business owner, the vendor and the decisions each system can make or influence.
  • Test the escalation route for ASIC's four situations. Run scripted conversations covering a complaint, a scam report, an insurance claim and a hardship request, and time how long each takes to reach a person. Fix any flow where a customer can't get through.
  • Put model and prompt changes under change control. Treat a new model version, a prompt rewrite or a vendor-side update like any other production change, with testing and sign-off. Firms running ITIL-based service management can extend their existing change process. Our ITIL 4 framework page covers how that process works.
  • Keep decision records you can explain. Log inputs, outputs, model version and any human override for automated decisions affecting customers. Check that the logs are kept for as long as your existing record-keeping obligations require.
  • Ask vendors direct questions. Find out how vendors notify you of model changes, what testing evidence they'll share and where customer data is processed. Write the answers into contracts at renewal.
  • Give the board a short AI risk register. List each system, its owner, its known failure modes and the date of its last review. The RBNZ, FMA and Commerce Commission all expect board-level visibility.

If you need help building the inventory or setting governance that will stand up to both NZ and Australian regulators, our AI consultancy team works with organisations on AI adoption and governance.

Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.

Get my free assessment How we help
Get new articles by RSS