Prime Minister Christopher Luxon is calling for new norms on how armies use artificial intelligence, but also says rules are "giving way to power".
At the same time, a leader of efforts to rein in military AI is accusing New Zealand of sitting on the sidelines when it used to lead.
Drones have become the biggest killer on Ukraine's battlefields, an AI targeting system is telling triggermen who to hit in Gaza, and in the US, the Pentagon has signalled its strongest embrace of generative AI yet.
Luxon told India's premier conference on geopolitics and geoeconomics the world was on the cusp of an AI transformation, but it could manage the risks.
"For example, militaries are already using AI, which means the international community is going to need to develop new norms about how this is done in a way that ensures compliance with the rules of war and ensures human responsibility in conflict," he said at the Raisina Dialogue.
But having just called for protective norms around AI, Luxon then noted - repeatedly - that the rules were being ripped up.
"We are seeing rules giving way to power ... rules are being undermined, whether those around territorial integrity, freedom of navigation, or laws of war ... the geostrategic picture I've painted is stark, rules are giving way to power," he said.
NZ Defence Force tests its new Vector Scorpion drones delivered in December. Photo: You Tube / NZ Defence Force
Years of international summits and declarations to try to set AI norms have not succeeded.
New Zealand's former disarmament minister Phil Twyford called the process "painfully slow", while MFAT on Wednesday said there had been "limited progress".
Oceania's only commissioner on a new global military AI forum, Australian Professor Toby Walsh, said the convergence of tech advances with the erosion of rules posed "a real risk of war becoming much more terrible".
"In Gaza and Israel ...AI is being used to decide who lives and who dies," he said.
Walsh said he had spoken to New Zealand diplomats, urging them to get the government to do more.
"It's surprising and disappointing" that it was not, he added.
The government defended its record, saying it was focused on efforts at the United Nations.
The first UN experts meeting of 2025 on killer robots early this month was still debating how international laws might apply, and what titles to put on five subject areas.
A key problem is that any new rules that constrain how AI can enable armies, might also constrain how AI for non-military advances, especially in developing countries.
More UN meetings on military AI are scheduled in September.
By contrast, what the US Army now calls 'HMI' - human-machine integration - is very directly accelerating.
Years ago, a 2018 NZDF report on the future of warfare described "battlefield targeting from satellites by artificial intelligence for killer robots", prompting the New Zealand Herald to comment, "Some of it sounds like the stuff of a sci-fi Hollywood blockbuster."
Yet this is exactly the technology the NZDF is helping the US Army test in Project Convergence this month.
Both politics and industry have increased the acceleration in recent months.
"The interesting development is not really what the Pentagon doing is, but the fact that Silicon Valley and venture capital is embracing this," said Walsh, author of Machines Behaving Badly.
The Pentagon's new AI-embedding project, Thunderforge, is a partnership not with huge defence contractors like Boeing and Lockheed, but with young San Francisco firm Scale AI, led by a chief executive in his late 20s.
The Trump Administration had signalled "a desire for nothing to hold them back", Walsh said.
"It does seem that there is a very large foot on the accelerator."
Servicemen from an artillery unit of the 128th Mountain Assault Brigade of the Ukrainian Ground Forces train to use a drone in Ukraine, on 28 December, 2023. Photo: Dmytro Smolienko / NurPhoto / NurPhoto via AFP
NZ appears to pull back
On the flipside, New Zealand appeared to have pulled back from the leadership it displayed for decades in getting rules on weapons strengthened, whether over nuclear weapons or mines, he said.
"My impression over the last decade of participating in this forum and United Nations ... is that New Zealand is less and less engaged with this problem."
Walsh was appointed a commissioner after two Netherlands-led summits on military AI - REAIM - in 2023 and 2024.
New Zealand sent just one person to the summits, compared to Australia's half a dozen.
This country did not endorse either summit's call or blueprint for action even though its Five Eyes partners did; it did not have anyone on the Global Commission on Responsible Military Artificial Intelligence that Walsh is part of.
"It was, you know, not a great ask to to sign up," said Walsh.
'New Zealand's voice is clear' - govt
The government pushed back, saying military AI initiatives had multiplied, so it was concentrating on UN efforts and a US-led 'Political Declaration on Responsible Military Use of AI and Autonomy'.
"New Zealand's voice is clear," said Associate Foreign Affairs Minister Todd McClay in a statement.
Associate Foreign Affairs Minister Todd McClay is defending the government's approach. Photo: RNZ / REECE BAKER
New Zealand had signed the the Bletchley Park AI safety summit declaration - though a year late in 2024 - and its successor summit, the Paris declaration, last month, although the Paris meeting carefully skirted military issues.
McClay has oversight because the coalition government did not appoint a Disarmament Minister, the second time a National-led government has downgraded the portfolio since it was set up after the nuclear-free legislation of the 80s.
An expert working group on killer robots set up in 2022 had just a few meetings, before MFAT pulled the plug last year.
Twyford, Labour's spokesperson on disarmament, said the government was not doing enough to "push back" against the big powers, the US, China, India and Russia that did "not want a bar" of military AI rules.
"If we don't, the consequences for the species could be horrific.
"I'm just not seeing evidence from the current government that they're making this a priority."
Phil Twyford, Labour's spokesperson on disarmament, says the government isn't doing enough to stand up to the major powers. Photo: RNZ / Samuel Rillstone
Twyford in 2021 got Cabinet to commit to supporting international rules. However, Cabinet caveats then carved out room for the Defence Force to remain interoperable with US and Australia's advances in AI technology.
Twyford and Walsh illustrate the splits that drag back the efforts on new norms: Walsh said the UN was ineffective, with any moves easily vetoed, but Twyford said the UN was the key forum to launch AI debates, that could lead to separate treaties that could not be vetoed, as happened with anti-personnel landmines in the 1990s.
Balancing the benefits with the risks
Luxon, at the Raisina Dialogue, mentioned the risks but stressed the benefits of AI as a whole.
"While we need to manage change, we cannot allow ourselves to be paralysed by the risks," Luxon said.
"For those who believe they can outcompete through this period of technological dislocation, the opportunities are there. The citizens, the companies, and the countries that embrace the coming change will be the ones that reap the dividends."
Christopher Luxon speaks to media in New Delhi. Photo: RNZ / Marika Khabazi
India hosts the next international summit on AI later this year.
Delhi does not support the negotiation of legally binding rules on autonomous weapons systems, and abstained on a related vote - draft resolution L77 - at the the UN last October.
Its defence force has just launched 75 AI-related technologies, as well as the Indrajaal autonomous drone security system that uses SkyCop drones to protect the likes of cities and power stations.
"Smartness" was as important for bombs as their "size and explosive capacity", its Defence Minister said recently.
Sign up for Ngā Pitopito Kōrero, a daily newsletter curated by our editors and delivered straight to your inbox every weekday.
