What parties must include in their AI policies, from data centre rules to independent oversight.
Commentary: So far, most parties have yet to announce major policies regarding AI.
The Green Party has called for a one-year pause on new AI data centres in New Zealand to allow time for regulations regarding environmental protection and energy consumption to be developed.
Labour has, perhaps, the most extensive AI policy to date, which includes elements such as establishing an “Office of AI” within the Department of the Prime Minister and Cabinet to coordinate AI policy across government, address how copyright material is used by AI tools, develop data privacy standards, and require data centres to establish their own renewable energy supply.
AI experts Dr Andrew Lensen, Christopher McGavin, and Dr Cassandra Mudgway authored an open letter to party leaders late last year that called for a “bipartisan effort to produce risk management-based AI regulation and to establish a national AI oversight body”.
I am a signatory of that letter.
Such a body, preferably an independent one, would be a basic first step to addressing the many potential harms of large language models and generative AI, and ensuring AI-powered tools and services are safe by design.
This body could also demand transparency from any organisation using AI, for example public information on what AI tools are being used and how, and how AI-powered decisions or outcomes can be challenged.
In terms of government use of AI, ideally each ministry or agency would designate someone as the “AI officer” who would be responsible for overseeing the responsible implementation of AI, act as a contact person if there were any problems, and coordinate with AI officers at other agencies to share information and best practices.
However, finding people with the necessary expertise to fill such a role could prove to be a challenge.
The harmful environmental and community effects of hyperscale AI data centres are well documented.
A blanket pause on building new AI data centres could have inadvertent negative consequences (such as data sovereignty), but communities should be provided with detailed, independently verified information on the impacts these data centres would have on their community and have more input before a data centre is established in their community.
Moreover, developing regulations to ensure hyper-scale AI data centres establish renewable energy supplies and use more environmentally responsible water-cooling systems would be a welcome step, something the Green Party has noted is currently not required.
AI researcher Jacob Coxon’s resignation from Anthropic in September, supposedly because “people building AI earnestly believe that it could kill us all by the end of the decade” certainly generated a lot of headlines.
However, it’s probably reasonable to treat that claim with a healthy dose of scepticism.
Several AI researchers and developers have made similar doomsday claims over the past few years including Nobel laureate and former Google employee Geoffrey Hinton, the “godfather of AI”. He resigned from Google in 2023 for similar reasons.
Coxon’s claim may have gained traction because of the revelation that autonomous AI agents from OpenAI managed to break out of a closed test environment or ‘sandbox’ to hack into systems run by another AI startup called Hugging Face to find solutions to a difficult evaluation task they were given.
Several experts and commentators have suggested that the announcement of the hack, claims about how newer AI models can result in human extinction, and recent calls for a ‘slowdown’ AI from AI company heads including Dario Amodei (Anthropic), Sam Altman (OpenAI) and Elon Musk (Grok) are all simply marketing ploys to tout how advanced their AI systems are.
However, the growing number of documented cases of AI agents breaking out of testing environments does raise some cause for concern.
Though these AI agents may not kill us all, they could cause havoc if they managed to break into banking systems, the electrical grid, or government and military systems and make changes before being found.
The recent hacking of Australia’s Medicare system by OpenAI bot agents is an indicative example.
Are international regulators and AI companies doing enough to regulate AI globally? In a word: no.
More specifically, there is no single global approach to the development or regulation of AI.
Most major AI companies are still going full steam ahead even while proclaiming their AI models may lead to harm.
Some jurisdictions, such as the EU, have put some statutory restrictions on AI development and use, though, even there, some of those regulations have been softened or rolled back in recent months.
Other countries, including Aotearoa New Zealand, have been reluctant to regulate AI to avoid limiting ‘innovation’ or perceived (though dubious) productivity and efficiency gains.
Dr Michael Daubs is a senior lecturer in the Media, Film and Communication programme at the University of Otago
This op-ed was first published on Newsroom on 2 October, 2026.