The Five Eyes agencies warn AI is reshaping cyber risk in months, not years. What it means for New Zealand organisations, and five questions for your board.
In late June 2026, the cyber security agencies of New Zealand, Australia, Canada, the United Kingdom and the United States put their names to a single statement. The Five Eyes agencies rarely co-sign documents, so a joint statement usually means the threat picture has changed faster than annual security planning can keep up with.
The line most people pulled out of it was this: "The rapid pace of frontier AI development means cyber risk assumptions can become outdated in months, not years."
What the agencies actually said
Underneath the framing, the statement makes three claims and asks for five things.
The claims are straightforward enough. AI lowers the barrier to entry, so more adversaries can do more damage. It compresses the time between a vulnerability being discovered and being exploited. This cuts both ways, because the same capability that helps attackers move faster can help defenders detect and respond faster, provided it is used deliberately and not just to save money.
The asks are not new, and that is rather the point. Reduce your attack surface. Patch faster. Retire or isolate legacy systems. Tighten identity and access. Assume breaches will happen and prepare accordingly. None of it would surprise anyone who works in security. What has changed is the cost of being slow about it.
The statement is also explicit that this is no longer an IT problem to delegate. Boards and executives are told to confirm that controls will hold up under real pressure, not just that the controls exist on paper. A tested incident response plan and a binder labelled incident response plan are different things, and you find out which one you have during the incident, when it is too late to fix.
How this looks from a New Zealand SOC
We run a security operations centre (SOC) delivering managed detection and response (MDR) across a range of New Zealand organisations, so the pattern the agencies describe is not a forecast we are just reading about. It is what comes through the ticket queue most weeks.
The window between disclosure and exploitation has been shrinking for a number of years, and AI accelerates it further. The comfortable assumption that you have a maintenance weekend to apply a critical patch no longer holds for internet-facing systems. The Five Eyes single out operational technology and long update cycles for exactly this reason.
The harder part sits under the identity recommendation. Most of the serious incidents we triage do not start with exotic malware. They start with a credential, a token, or an over-permissioned account that nobody had looked at in eighteen months. AI makes phishing and account takeover cheaper and more convincing at scale, so the gap between strong identity controls and weak ones widens quickly. The customers that come through cyber incidents in good shape tend to be the ones that had already done the boring work on multi-factor authentication, conditional access and permission reviews.
Five questions worth taking to your next board meeting
The Five Eyes agencies framed this as a leadership responsibility, so here are the questions we are putting to the boards and executives we work with.
- Do we have evidence our controls work, not just a list of them? When did we last test patching, identity, monitoring and incident response under realistic conditions, and what did the test find?
- How fast can we detect and respond? Hours, days, or do we not know? If the honest answer is that we are not sure, that uncertainty is the first thing to fix.
- Where is our legacy and internet-exposed surface? Which unsupported systems sit on easy paths, and is there a funded plan to isolate or retire them, or just an intention?
- Are our cyber leaders resourced to act at the required pace? Do they have the authority, budget and access to move, or do they carry the responsibility without the means?
- Are we governing our own AI use, both ways? Are we using AI to strengthen defence, and have we assessed the AI our staff and vendors are introducing into the supply chain?
What we would prioritise first
If a New Zealand organisation read the statement this week and asked us where to start, we would not hand back all five actions at once. We would sequence them.
Start with identity, because it is where most real incidents begin and where the controls are mature and affordable. From there, close the gap between vulnerability disclosure and patch on any system exposed to the internet. Legacy and exposed attack surface comes next, and that one is usually a budget and prioritisation conversation rather than a technical problem. Detection and response capability and a tested incident plan run alongside all of it, because breaches will happen, and the difference between a contained incident and a crisis is how fast you move in the first hour.
The agencies closed by saying success will not come from owning the most tools, but from getting the basics right, fast. That is true, as far as it goes. But the basics are only worth anything if someone has tested that they hold under pressure. For most boards, funding that testing is the thing to sort out before almost anything else.
Fenrir Security runs a managed detection and response service for organisations across New Zealand. If the five questions above are harder to answer than they should be, we can help you work through them. Get in touch.
Source: Five Eyes cyber security agencies statement, "The AI shift in cyber risk: why leaders must act now," 22 June 2026, signed by the heads of the cyber security agencies of New Zealand, Australia, Canada, the United Kingdom and the United States. The NZ signatory is Catriona Robinson, Head of the National Cyber Security Centre (GCSB). The statement is published by the NCSC and partner agencies, including CISA and the Australian Cyber Security Centre.
