Most of the AI news we cover is about what your team can do with it. This week the more useful story is what attackers are doing with it, and how little time that leaves a business to react.
On 26 October 2026, the Australian Signals Directorate released its Annual Cyber Threat Report for 2025–26. It describes AI as “the defining force in the cyber threat landscape” over the past year. New Zealand’s National Cyber Security Centre made a similar call in its own report earlier in October. For owners, finance leads and operations managers, the practical message is simple: the gap between a weakness becoming known and someone using it is shrinking, and the routines most businesses run were built for a slower pace.
What changed
Australia’s report. According to Cyber Daily’s coverage, Australia’s cyber agency responded to more than 1,500 incidents over the year and notified organisations of potentially malicious activity more than 4,400 times, about 80% more than the year before. It says AI helps attackers find and exploit weaknesses faster, and urges organisations to fix vulnerabilities quickly and prepare for service disruption. The average reported cost of cybercrime to a business was A$93,500.
New Zealand’s report. NCSC NZ’s Cyber Threat Report, released around 5 October 2026, recorded 369 incidents of potential national significance in 2025/26, 162 of them linked to criminal or financially motivated actors, up 18%. It warns that advanced AI tools now mostly limited to leading models could reach malicious actors by early 2027. Its head, Catriona Robinson, put it plainly: “Government cannot protect every organisation from every cyber threat.”
A real example of the speed. On 21 October 2026, Microsoft attributed an attack involving ConnectWise’s Cloud Backup product, which protects Microsoft 365 and Google Workspace data, to a China-aligned group it calls Storm-0940. The attackers reused stolen credentials belonging to ConnectWise cloud apps, including the backup app, and an open-source tool that coordinates a team of AI agents. In several cases they had privileged access less than three hours after first targeting an organisation, though Microsoft says most of the targeting was unsuccessful. Microsoft noted that agents can work around the clock, including weekends and holidays.
That last example matters for New Zealand businesses because of how it got in: not by breaking a password on a staff account, but through a trusted app that already had wide access to company data.
Why it matters for how teams work
Monthly patching is too slow for exposed systems. Many businesses update software on a monthly cycle, or when the IT provider gets to it. Australia’s Essential Eight framework already sets a 48-hour window for critical vulnerabilities in internet-facing systems, or where a working exploit exists. If attackers can move from a public flaw to an attack in hours, that kind of target needs to be the norm for anything reachable from the internet.
Connected apps are now a front door. Backup tools, reporting add-ons, email signature tools, AI assistants and integrations often hold broad, long-lived access to your Microsoft 365 or Google tenant. Staff accounts have MFA and get reviewed when people leave. Apps rarely get the same attention.
Out-of-hours is when attacks land. An automated attacker doesn’t wait for Monday. If nobody would see an alert until the next working day, assume the damage window is the whole weekend.
Fraud gets more convincing. NCSC NZ notes AI is already being used in phishing and scams. A well-written email that matches a supplier’s tone is no longer a sign it’s genuine, so payment controls have to rely on process, not judgement.
How NZ businesses could respond, by sector
- Distribution and wholesale: treat bank detail changes from suppliers as high risk. Require a call-back to a number already on file, and a second person to approve, before any change goes live.
- Construction and trades: job management apps, subcontractor portals and shared drives often connect to email and files. Review which apps have access and remove ones tied to finished projects.
- Manufacturing: separate the networks that run plant equipment from office systems where you can, and make sure a ransomware incident in the office can’t stop production.
- Agriculture: many operations rely on one person and one laptop. Make sure backups exist outside that laptop and that MFA is on for banking, email and farm management software.
- Professional services: client files are the asset. Check who can share externally, and whether your backup provider’s access to your tenant is limited to what it needs.
AI can help on the defence side too. A business on Microsoft 365 can use the AI tools it already pays for to summarise security alerts, draft a plain-English incident plan, or review its list of connected apps. Treat the output as a starting point that someone qualified checks, not as the control itself.
Risks, costs and governance
Privacy Act obligations. Under the Privacy Act 2020, a breach where it’s reasonable to believe someone has suffered or is likely to suffer serious harm is a notifiable privacy breach. The Privacy Commissioner expects it to be reported within 72 hours of you becoming aware it’s notifiable. If a connected app or IT provider holding your customers’ information is breached, you may still need to assess and report it, so know who you’d call and who decides. This is general information, not legal advice.
Your IT provider is part of your attack surface. Remote management and backup tools give providers powerful access. Ask your provider how that access is protected and how quickly they patch their own tools.
Faster patching costs something. Shorter windows mean more out-of-hours work and occasionally a broken update. Agree the trade-off in advance: which systems get patched within 48 hours, and who accepts the risk if one can’t be.
Human sign-off still matters. Automation can block a suspicious sign-in or isolate a laptop. Decisions to pay, restore or tell customers need a named person.
What managers should do differently
Stop treating cyber security as an IT task that runs in the background. The decisions that matter here are management ones: how fast is fast enough, which apps are allowed into company data, who is on call, and what the business does if systems are down for three days. Put a named owner on each, and review them at least twice a year, not only after an incident.
Ask for numbers you can track: how long critical patches take to apply, how many connected apps have access to your tenant, and when you last tested a restore from backup.
Try this in the next 30 days
- List every system reachable from the internet, including firewalls, remote access, websites and portals, and confirm who patches each one.
- Agree a patch target, such as 48 hours for critical fixes on those systems, and ask your IT provider to report against it.
- Review connected apps in your Microsoft 365 or Google admin centre. Remove anything unused and reduce permissions on the rest.
- Ask your IT provider how their remote management and backup tools are secured, and whether they were affected by recent incidents.
- Check out-of-hours cover. Who sees a critical alert at 2am on a Saturday?
- Test a restore of one important system or folder from backup.
- Write a one-page incident plan: who decides, who you call, and how you’d assess a notifiable privacy breach.
- Brief finance staff on call-back checks for any change to payment details.
Where to start
Start with the connected apps and internet-facing systems, because that’s where speed matters most. If nobody in your business clearly owns security decisions, our fractional CTO work includes security reviews, and our AI and agents work covers the access your AI tools have. Or get in touch for a conversation about where you stand.