When the NSA’s classified messaging system went down last week, everyone assumed foreign hackers or a rogue AI. The cause was duller and, for business, more worrying. US Director of National Intelligence Jay Clayton said the outage came from an upgrade responding to vulnerabilities in the PHP programming language. It was a fix, not an attack.
That is the lesson New Zealand business leaders need to take on board. The next time email, payroll or a patient management system falls over, the likeliest culprit is not a criminal offshore. It is a vendor patching in a hurry. Outages have stopped being freak events and become a recurring cost of running software.
AI found the holes faster than anyone can fill them
The acceleration traces back to March 2026, when Anthropic shared its Mythos model, which is highly capable of scanning code for bugs, with a select group of tech and security firms. They immediately found hundreds of vulnerabilities, some dormant for decades.
The numbers since are extraordinary. Microsoft patched 3,554 security flaws in the 2.5 months before its major November outage, more than in 2024 and 2025 combined. Across the industry, more than 100,000 confirmed vulnerabilities have been found this year, more than double 2025’s 48,000. Twenty chief executive Joe Lin describes a collision between “the speed with which we are discovering new vulnerabilities” and “our ability to address them”.
The past week’s casualties included Microsoft Outlook, US air traffic control, JPMorgan Chase, United Airlines, Apple and Workday. Former CISA director Chris Krebs said all the available evidence suggests the events are not connected. US Representative Jim Himes went further, warning that the pace of discovery and patching “could cause more frequent outages”. Retired General Paul Nakasone, former head of the NSA, put it bluntly: “what we are seeing is a new normal”.
Call it a change problem, the bill is the same
The industry is arguing over blame. Azul CTO Gil Tene says there is “strong reason to believe” AI-driven patch velocity is already causing major incidents. Filigran CEO Samuel Hassine counters that these are updates “pushed onto systems that can’t take much change”.
For a business owner whose systems are down, the distinction is academic. Both diagnoses point to the same conclusion: brittle systems are being asked to absorb more change, faster, than they were built for. That describes a good chunk of New Zealand’s corporate IT estate.
New Zealand has already had three warnings
The health sector has suffered Microsoft-linked disruption three times in 2026: January in the upper North Island, July in Waikato, and again this month. Cogent IT expert Tony Grasso is scathing about the orthodoxy behind it. New Zealand decided “servers are bad, because they cost money, and the cloud is good”, he says, “but then when Microsoft has that problem you’ve got no way to fix it.”
Unions want hospital IT treated as part of life saving clinical care rather than relying solely on Microsoft, while Health NZ director James Allison insists disruption was minimal, pointing to Starlink and satellite phone contingencies.
This isn’t an argument against the cloud. For most firms it remains cheaper and more secure than a server cupboard. But the move was sold on cost savings, and almost nobody priced the vendor itself as a single point of failure. If the most risk-averse sector in the country got caught out three times in a year, most SMEs are in no better shape.
Regulators see the risk, the rules arrive in 2028
On paper, the local picture looks benign. The NCSC recorded 4,673 incident reports and $23.8 million in direct losses in 2025/26, down from 5,995 and $26.9 million the year before. But the agency warns that by early 2027 malicious actors may gain AI capabilities currently confined to frontier labs, with attacks unfolding “often faster than businesses can detect and respond”.
The Reserve Bank now names concentration risk in shared cloud, software and AI providers as a financial stability concern, because one disruption can hit multiple institutions at once. Yet its Operational Resilience Standard does not take effect until 2028. In the meantime, the RBNZ and FMA are urging firms to “get the basics right” now. That leaves a two-year gap where the exposure is live and the formal obligation isn’t. Waiting for the rulebook is the wrong call.
Budget for the outage, not the headline
The fix is unglamorous. Mitiga field CISO Brian Contos recommends staged rollouts, monitoring and tested rollback procedures. For business leaders, that translates into a short list: map which vendor sits under each critical function, build an offline fallback for the processes that cannot stop, and ask suppliers hard questions about how they stage and reverse updates.
The patch wave won’t ease while AI keeps finding flaws faster than humans can fix them. Firms that treat outages as a line item in 2027 will ride out the next Microsoft wobble. Those still waiting for a freak event will find it has become a fixed cost.
Sources
- NZ Herald: ‘New age of digital chaos’: Welcome to the age of endless digital breakdowns (2026-11-22)
- NZ Herald: What we know about why everything seems to be breaking at once (2026-11-18)
- NBC News: The AI race to fix security flaws could be making tech outages more common (2026-11-21)
- SecurityWeek: Industry Reactions to Major Outages and AI Patching Debate: Feedback Friday (2026-11-20)
- RNZ: Unions, IT expert warn of health sector’s reliance on Microsoft after outage (2026-11-20)
Join the discussion
Add useful context, ask a good question, or challenge an idea — keep it specific and respectful.