Ernst & Young got hit. ShinyHunters — the same extortion group that went after Vercel, Instructure, and Charter Communications — has claimed the breach, and the July 31 deadline they gave EY to negotiate has passed. The stolen data includes client tax documents: names, addresses, Social Security numbers, financial account numbers. EY is offering two years of credit monitoring, which is what companies say when the horse has already bolted.
The breach itself is straightforward. Someone compromised a third-party IT support platform that EY used for tax work. The attacker had access from March 28 to April 12 — two weeks of downloading client documents from support tickets. At least 1,366 US residents are confirmed affected so far, but EY’s global client base means the real number is likely much higher.
This is a supply-chain attack. Not a direct hack of EY’s systems, but a compromise of the software EY depended on. Same pattern as the Shai-Hulud npm worm that hit 868 packages with 2 billion monthly installs. Same pattern as the Vercel breach where ShinyHunters got in through a compromised OAuth app. The vulnerability isn’t your code. It’s everyone else’s.
Here’s the part nobody wants to hear
The IT community has been having a quiet conversation about isolation. Not the “segment your network” kind. The physical kind. Air-gapping.
For years, air-gapping was something nuclear facilities did. SCADA systems. Military networks. The idea was simple: if a machine has no connection to the internet, no remote attacker can reach it. It was considered overkill for normal businesses. Connectivity was the default — everything online, everything API-accessible, everything cloud-connected.
That assumption is breaking down. Here’s why.
AI changed the speed of attacks. Anthropic disclosed in November 2025 that a China-linked group called GTG-1002 used Claude to run a cyber espionage campaign against roughly 30 global targets. The AI executed 80 to 90 percent of the operation. Humans stepped in at four to six decision points. Anthropic called it “the first documented case of a large-scale cyberattack executed without substantial human intervention.” CrowdStrike’s 2026 report shows AI-enabled adversary attacks up 89 percent year over year.
The breakout time is now 29 minutes. That’s how long it takes an attacker to move from initial access to lateral movement inside a network, according to CrowdStrike. In 2022, it was over 8 hours. Human SOC teams cannot respond in 29 minutes. By the time someone sees the alert, the attacker is already in the next system.
AI-generated phishing clicks at 54 percent. Human-written phishing manages 12 percent. KnowBe4 found that 82.6 percent of phishing emails analysed in late 2024 and early 2025 used some form of AI-generated content. The lures are better. The volume is higher. The defenders are the same.
1 in 6 breaches now involve attacker AI. IBM’s 2025 Cost of a Data Breach Report put a number on it. Not a projection — measured. The average breach cost in the US is $10.22 million.
The pattern across all of this: attacks are getting faster, cheaper, and more automated, while defence remains manual, slow, and expensive. Something has to give.
What isolation actually means in 2026
Nobody is suggesting every server goes offline. That’s not practical and never will be. But the conversation is shifting toward selective isolation — keeping critical systems on networks that simply have no path to the internet.
Backups. The most obvious case. Ransomware can’t encrypt what it can’t reach. Air-gapped backups — stored on drives that are physically disconnected — are the one recovery method that works regardless of how sophisticated the attack is. This is already best practice. The question is whether more companies will actually do it, given what’s happening.
Operational technology. Factory systems, power grids, water treatment. New Zealand’s National Cyber Security Centre already publishes guidance on isolating vital OT systems. The NCSC’s advice is direct: “Isolating vital OT and enabling systems can disrupt the ability of malicious cyber actors to achieve their goal, contain active incidents, and allow for safe rebuilding of compromised systems.” That guidance exists because the threat is real.
Sensitive data stores. Tax records. Health records. Financial systems. The EY breach happened because client tax documents were sitting in a support ticket system that was connected to a third-party platform. If those documents had been on a system with no internet path, the attack wouldn’t have worked. That’s not a sophisticated defence. It’s a blunt one. But blunt works when the attack is blunt.
Development environments. The Shai-Hulud npm worm spread because developers ran npm install on compromised packages, which then scraped credentials from their machines. If build systems for critical software were isolated from the public npm registry — with packages vetted and transferred manually or through a one-way diode — the worm would have stopped at the first machine.
The trade-off
Isolation is expensive. It makes updates harder. It breaks automated testing. It slows everything down. The LinkedIn discussion around air-gapped security in 2026 calls this the “Air-Gap Paradox”: the systems that most need isolation are also the hardest to maintain when isolated.
But the calculus is changing. When a breach costs $10 million, when an AI can scan 40,000 vulnerabilities and exploit 56 percent of them without authentication, when the attacker moves from access to lateral movement in 29 minutes — the cost of isolation starts looking like insurance. Not the kind you buy and hope to never use. The kind you buy because the event is becoming likely.
New Zealand
New Zealand’s Cyber Security Strategy 2026-2030, published in February, explicitly addresses critical infrastructure protection. The NZ NCSC has isolation guidance available. The question for NZ businesses is whether they’ll treat it as something only large enterprises need, or whether the EY breach — a Big Four accounting firm, supposedly among the most security-conscious organisations on the planet — changes that perception.
If EY can get hit through a third-party platform, any NZ firm using SaaS tools can get hit the same way. The difference is that NZ firms generally don’t have EY’s resources to recover. For a New Zealand accounting firm or law firm holding client tax records, a breach like this could be terminal. Not $10 million in costs — just the loss of trust.
Isolation isn’t a panacea. Stuxnet proved that even air-gapped systems can be reached. But it raises the cost of attack dramatically. And right now, attacks are cheap. AI makes them cheaper. Making them expensive again might be the only sustainable defence.
— CJ Murden, editor of Singularity.Kiwi. Former digital technologies teacher, author of AI-focused books. Writing with a New Zealand focus.
