Digital oligarchy: Can US tech giants override NZ sovereignty?

Sunday, 26 April 2026

ANALYSIS: People deserve to know what government agencies are using artificial intelligence for, and a public register could be the answer.

The call for greater government transparency on AI was made at the University of Auckland’s two-day ALTeR artificial intelligence conference at the university’s Business School this month.

The event was sharing the venue with the cast and crew for Disney movie Zombies 5, but it was other monsters that haunted the gathered tech experts; the trillion-dollar US tech companies some felt threatened New Zealand sovereignty.

There was an “extreme concentration” of AI power in the hands of companies owned by men who could be recognised by their first names, and who were not to be trusted with it: Elon, Mark, Sam, Jeff, and others.

Read more:

• Worshipping at the touring evangelical church of the Microsoft AI Tour

• Singaporean tech giant defends Southland ‘AI factory’ promised to add billions to GDP

• The Office Blueprint: ‘AI brain fry’ a real risk for office workers in the push to productivity

Would the “tech bros”, as they are known, and their US political masters in the White House, end up with “kill switches” which they could use to threaten and control the New Zealand Government?

And would the White House’s insistence that US tech companies must submit to it using their systems for “all lawful uses” give US spies access to New Zealand data hoovered up by those companies’ AI systems?

These were uncomfortable questions for which there were no definitive answers at the conference.

But Marta Andhov, associate professor from the University of Auckland’s ALTeR Centre (ALTeR stands for Advancing Law and Technology Responsibly), asked: “Do we really want to have all our systems running on just a couple of large companies’ infrastructures?”

New Zealand’s political leaders have been slow to develop AI governance rules and laws, the conference heard, and was the last country in the OECD to publish an AI strategy, which it did in July last year.

It is a self-defined “light-touch” strategy with New Zealand as a “fast adopter” of AI focused on economic efficiency, and the finger-in-the-air hope of adding $76 billion to New Zealand GDP by 2038 by enhancing productivity in agriculture, healthcare, education, tourism, and government.

However, there was a flip-side to that rosy scenario, the conference heard, which threatens high unemployment, puts our data in the hands of foreign actors, erodes human capacity to think critically, and erodes the tax base, as untaxed AI displaces income-tax generating human labour.

Professor Jeannie Patterson from Melbourne University’s Centre for Artificial Intelligence and Digital Ethics, told the conference: “The downside is we are all slaves to three big companies.”

Dependency was a real issue with AI, both for individuals, and governments.

“It's like sugar, and people get addicted to it,” said US expert Jessica Tillipman from George Washington University Law School.

Harm and lack of transparency

The conference heard New Zealanders needmore transparency from government agencies over the deployment of AI applications with power over their lives.

A public register of AI applications in the public service, with descriptions of how it operated, and what security assurances were in place, which could bring transparency, argued Professor Ali Knott from Victoria University of Wellington.

A recent use of AI by ACC to identify long-term claimants to “exit” from ACC’s pool of 25,000 long-term injury sufferers was a local example of transparency failure presented to the conference.

Professor Alexandra Andhov from the ALTeR Centre said ACC had deployed AI, or some other machine-learning system, to identify cases that may benefit from “further review” by human staff members.

Removing claimants from the long-term pool would save ACC money, she said.

But Official Information Act requests had been needed to reveal how the system was working.

“In the year to June 2025, a report of 8000, so out of those 25,000, 8000 long-term claimants were removed,” she said.

However, it appeared people were being deemed capable of work, who really were too badly injured to do it, putting them through needless cost and stress.

“Nearly 2000 people were exited and subsequently re-entered into the long-term pool,” Andhov said.

Human harm through automation

Horror stories of the reality-warping power of automated systems were cited, showing how human beings could become victim to technologies over which humans failed to maintain competent oversight and control.

There were three global marquee failures.

One was the British Post Office scandal in which a faulty accounting system led to hundreds of sub-postmasters being wrongly prosecuted for theft and fraud, with around 230 wrongly imprisoned, hundreds bankrupted, and more than a dozen committing suicide.

The second was Australia’s “Robodebt” automated debt assessment and recovery system, which wrongly identified thousands of people as having been overpaid benefits, and hounded them for repayments, driving some to take their own lives.

The third was the Dutch childcare benefits scandal in which around 26,000 parents were wrongly accused of making fraudulent benefit claims, resulting in demands to repay their allowances, driving many into financial hardship, and leaving some homeless.

“These technologies are often presented as faster, cheaper, and more efficient,” Marta Andhov said. “Sometimes they are, but those benefits do not happen automatically. We already have seen many cases where technologies introduced into public administration have caused real harm.”

Automation could very rapidly scale up harm.

In the case of the Robodebt failure, she said: “People had just 21 days to prove that [the notices] were wrong, and notices ballooned from 20,000-a-year to 20,000-a-week.”

These were glimpses into a world in which, as one delegate to the ALTeR conference put it: “We are devolving our critical thinking to AI models.”

Code has no conscience

Code has no conscience, so humans must be the conscience of AI, the conference heard.

Some felt New Zealand was too small to regulate the largely unregulated US AI companies.

Netsafe’s Michael des Tombe spoke about Meta, which slashed human moderators looking for harmful content after Donald Trump became US president in 2025, and is accused of developing addictive services.

“Nobody has cracked the nut of how you regulate Meta, which is not based in your territory,” he said.

“I don't know the answer to that,” he said. “If anyone in the room does …”

That question brought a call of “I do. Stop f..king paying them.”

That was from Don Christie, boss of large, successful New Zealand-based developer Catalyst, which develops services for schools, universities, and governments, using open source software that is not controlled by giant tech companies.

Catalyst had its own set of AI principles that it built applications with, contrasting to the lack of them in the government’s AI strategy.

“People are always in control of our AI systems, which we design to benefit people and minimise harm,” they say. “If an AI system’s decisions can have adverse impacts on people, there must be staff who can review and override those decisions.”

Catalyst’s principles dictate that in every use: “We outline the likely impacts of risks. We always clearly indicate when AI is in use. We describe in plain language what AI is doing. We indicate to people using a system enhanced by AI, whether their data is used for AI training purposes.”

The Australian government had a set of AI ethics principles, the conference heard. The New Zealand government did not.

Breaking up with the Tech Bros

But is New Zealand really powerless?

The country may have failed in its “Christchurch Call” following the live-streaming of the terrorist attack on two mosques in March 2019, but Christie’s interjection called for the government to use its buying power to push for change.

Government procurement amounts to around $50 billion each year, and a growing proportion of that is expected to be for AI applications.

Alexandra Andhov said: “This is the lever most directly in any country’s hands. And definitely needs to be in New Zealand.

“What governments buy, and the standards they require of suppliers, is sovereignty expressed through contract. We might not be fully capable of regulating Silicon Valley from Wellington. But we can decide what our hospitals, schools, universities and courts are to run on- under whose law, with what conditions, with what liability when the architecture fails.”

It does appear the government is trying to get smarter.

Aware that individual government agencies lacked AI knowledge, the government has set up the Government Chief Digital Office to centralise, and build, capability, though we are behind Australia on things like model AI clauses in procurement.

The GCDO’s latest newsletter suggests some data and digital security concerns are being considered, even if the results are not communicated well enough to allay the fears of the experts at ALTeR.

Some at the conference felt New Zealand should develop its own AI applications even at the risk of them being more expensive, and less powerful, than those offered by US big tech.

New Zealand was never going to be able to develop its own AI large language models alone, but some at the conference called for it to work with Australia and Europe, which are far more assertive about AI controls.

Digital sovereignty is not a pipe-dream to some countries in Europe and Asia.

Denmark, which US President Donald Trump has threatened with the annexation of Greenland, is phasing out Microsoft products in favour of open-source alternatives like Linux and LibreOffice to reduce its reliance on US tech firms and secure its data.

As Marta Andhov said, “We are seeing some European countries breaking up with the US products, such as Microsoft.”