Summary

  • RBNZ and FMA have made frontier-AI cyber exposure a clearer supervisory issue, with attention to board oversight, shared suppliers and recovery readiness.
  • The agencies say their expectations build on existing duties; the announcement does not establish a standalone AI rule or identify a firm that failed them.

On 17 September, the Reserve Bank of New Zealand published three possible futures for banking in 2035 and said the study did not signal new policy or supervisory settings. Four weeks later, the Reserve Bank of New Zealand – Te Pūtea Matua (RBNZ) and the Financial Markets Authority – Te Mana Tātai Hokohoko (FMA) moved from that horizon-scanning frame to present-day supervision: their 15 October statement sets out expectations for cyber risk associated with frontier artificial intelligence.

The distinction is a matter of regulatory status. The study explored how AI, cloud services, data sharing and interconnected providers might reshape banking. The joint statement says regulated entities should manage current exposures under existing prudential, conduct, cyber-security and operational-resilience obligations. It does not announce a new statute, a separate AI compliance regime or enforcement action against a named institution.

The expectations were outlined by Angus McGregor, RBNZ Assistant Governor for Financial Stability, and Louise Unger, FMA Executive Director for Response and Enforcement, at a roundtable co-hosted by RBNZ, FMA, the Department of the Prime Minister and Cabinet, and the National Cyber Security Centre. The agencies describe their roles separately: the Reserve Bank approaches cyber risk through financial stability; the FMA through conduct. A shared discussion venue does not merge those mandates.

The officials group the challenge into three connected areas: higher cyber risk as AI capabilities advance; pressure on governance and oversight; and growing third-party and concentration risk. The National Cyber Security Centre’s September threat report says advanced models can accelerate vulnerability discovery and reconnaissance and make some attacks, phishing and deepfakes easier to produce. That is threat context, not evidence that a New Zealand financial firm has suffered an AI-enabled breach.

For institutions, the first test is whether leadership can see where AI is used and what changes it makes to the organisation’s risk profile. The speech calls for a proportionate approach aligned with each firm’s business, risk appetite and resilience objectives. Boards and senior managers should understand material AI uses and dependencies, set appetite and accountabilities, retain enough expertise to challenge management, and monitor changes in control weaknesses, incidents and exposure. Independent assurance may test whether controls work as intended.

The technical baseline remains familiar: strong identity and access management, current asset records and timely patching. The speech says AI can shorten the time between finding a vulnerability and exploiting it, especially in unsupported or legacy systems. Faster, risk-based remediation may be necessary, but speeding up patches can itself create operational risk; the supervisor’s message is to manage both sides of that decision. Firms should also ask whether manual monitoring and triage can keep pace, and govern any AI or automation they use for defence.

The third-party question reaches beyond a supplier’s contract label. Financial institutions are asked to understand how providers use AI in their own environments, what threat exposure that creates, how resilient key suppliers are, and whether a small number of providers could affect many firms at once. Contracts should support timely notice of material changes, vulnerabilities and incidents. Responsibilities may be shared with a provider, but the regulated entity remains accountable for its operational risk.

The speech also moves from prevention to recovery. Firms should assume that incidents will occur and test severe but plausible cases such as a high-speed attack, a zero-day vulnerability, a model failure or an outage at a critical provider. The exercise should reveal whether manual workarounds, backups, failover and recovery plans can sustain important services; critical suppliers can be included where appropriate.

This builds on an established supervisory line. RBNZ’s 2021 cyber-resilience guidance already assigns ultimate cyber-resilience responsibility to boards and covers third-party management. FMA’s 7 October update describes continuing sector surveys of operational-resilience maturity, while the September banking study remains explicitly exploratory. The October speech makes the frontier-AI threat scenario more specific within that framework. Its practical effect will be tested through supervisory engagement, not inferred from the roundtable itself.

The public documents name no financial institution, AI vendor or measured concentration level. They do not establish that any firm has weak controls or that an AI incident has occurred. They do make the expected evidence clearer: a board should be able to connect AI use, supplier dependencies and cyber scenarios to named owners, tested controls and a credible recovery path.

Sources