Using AI to streamline public services and automate decision-making must be done carefully. If it isn't, our people, our communities, our values, and our country could be harmed:
benefits payments denied to someone who desperately needs them;
advice to ministers that outsourced the thinking to overseas Big Tech companies;
government initiatives that quietly forget our unique economic, legal, cultural, historical, social, and environmental context;
taxpayer dollars sent overseas as AI providers extortionately raise the subscription price;
New Zealanders' sensitive data shared with shadow AI;
a government that is not equipped to work for us.
We, the people of New Zealand, shouldn't be asked to blindly trust Artificial Intelligence (AI) and Automated Decision Making (ADM) use in government. We should be shown that it is trustworthy. And if it's not worthy of our trust, we should be given the information to recognise that, and do something about it.
This petition asks for government use of AI to be transparent to the public, have independent oversight, and be guided by what the public thinks is acceptable.
This will enable us to protect ourselves, our communities, and our unique country from bring harmed by AI use in government.
That the House of Representatives legislate to require:
1. a public register of government AI systems’ function, performance, and impacts;
2. formal oversight mechanisms for government AI use, proportional to risk; and
3. involvement of the public in a deliberation process to define acceptable risk and use of AI in government systems.
Official petitions to Parliament can only be 300 characters long. This website gives more information about what we're asking for, and examples of what it might look like.
We can't ask questions about these AI systems if we don't know they exist! A centralised, plain-language public register means that we can know when and how AI is affecting our lives. If a machine informs a decision about our welfare, tax, or legal rights, that system's functionality and performance should be a matter of public record. See this page for more detail about this idea.
Decisions made by government agencies matter because they can have a big impact on our lives. We need to understand how AI will affect these decisions before the systems are launched. We also need ongoing monitoring after the systems are launched to ensure they are still working as intended.
New Zealand has some frameworks and toolkits already, but these are voluntary, and it's unclear if they're being used. For example, since it was introduced in 2023, no examples of an Algorithm Impact Assessment have been published. This type of due dilligence should be required, and someone should be checking whether it's being done.
That being said, some uses of AI aren't very risky at all. The strength of the oversight should be proportionate to the level of risk. No one is worried about spellcheck, but we are definitely worried about using live facial recognition and automated decision making. See this page for more information about this idea.
What uses of AI are the New Zealand public okay with? What risks are people most worried about, and what trade-offs are people willing to make to get the benefits that can make our public service flourish?
We don't know, because the government hasn't asked us. We haven't had the sort of national conversation needed to understand what New Zealanders really want. We need dedicated resources for public engagement and consultation. This will help to align the government's use of AI with Aotearoa New Zealand's unique economic, social, cultural, geopolitical, and environmental contexts.
If AI is the defining technology of our time, then all of us should have a say in how it is used. See this page for more details about this idea.
This petition is not calling for a ban on the use of AI by the public sector. AI is a powerful technology, but it's not perfect today (and it may never be). We want trustworthy use of AI - that is, AI built responsibly and used in safe ways. It must be AI that is proven to be worthy of our trust. Having the right transparency and oversight mechanisms in place, informed by the public's views about AI, gives the public sector the right boundaries to use AI with confidence and public support.
AI in the public service is growing rapidly: in a 2025 survey, 70 agencies reported 272 use-cases — a large increase from 37 agencies reporting 108 uses in 2024. It's important that these agencies are responsible with our data, and carefully consider the impact that AI use could have on our country and our people.
81% of New Zealanders believe that AI regulation is required, and only 23% believe current safeguards are sufficient, according to a KPMG survey in 2024/2025.
Let's make this too big to ignore.
Share this page or the petition with three friends who value open, accountable government.
Start a national conversation about what it would mean for AI to improve government services, and not erode them.
Get 10,000 to 100,000 signatures before the petition closes on 30 September.
Share our message with candidates in the upcoming election.
Present the petition to Parliament in early 2027.
Achieve legislative change and funding to build trust in government use of AI—by making sure that it is worthy of our trust.
Adopting AI for government processes should require proper consultation, careful analysis, and a well-defined problem. The AI should be tested to ensure that it works well for our context. Is it trained on data that represents New Zealanders? It is systematically biased? Will it reproduce historical inequalities? Does it align with the values, culture, and priorities of the people of New Zealand? Will the humans interacting with the AI system over-rely on the outputs, and therefore think less critically about their work? Does the AI's output impose an imported approach onto our bicultural, unique, and diverse nation?
All of these questions should be considered before deciding whether to use AI for a specific task, and exactly how it should be used.
AI could make public services more efficient, but it's no point moving fast in the wrong direction.
New Zealand's current approach to AI governance isn't great:
New Zealand only scored 27 out of 100 in the Global Index on Responsible AI in 2024. This just isn't good enough.
Australia scored 56. They've even improved since then. In 2025, they made it mandatory for public service departments to share how they're using AI.
While 70 public sector agencies said they were using AI in 2025, it's unclear who these agencies are, and whether they are using AI responsibly. In contrast, there are only 30 signatories on the Algorithm Charter, which means only 30 government agencies have made a commitment to responsible deployment of algorithms. Even then, there is no oversight as to whether or not they are upholding those commitments.
We deserve to know when AI is being used in government, how well these systems work, how they're expected to impact us. This will give us, the people on New Zealand, the information we need to challenge decisions that have been made unfairly.
In principle, we could get some of this information using the Official Information Act. But that would be inefficient for everyone, and the OIA requests don't produce documents if those documents don't exist. If AI is impacting our country and our lives, this information should be available to the public and easy to find.
New Zealand government agencies are rapidly adopting automated decision-making (ADM) and artificial intelligence (AI) tools. Some of these uses are disproportionately likely to impact our most vulnerable:
automatically reducing a benefit payment without confirming the recipient has a chance to update their information;
using generative AI to transcribe police notes, with a 1% hallucination rate "including making up racial commentary, violent rhetoric and imagined medical treatments"
entering New Zealanders' private and sensitive data, such as taxpayer information, health information, or interactions with police into free AI models that have unclear privacy and security measures.
All of these uses have the potential to create real-life physical, financial, legal or psychological harm to New Zealanders.
Overseas experiments with these technologies have shown what can go wrong when AI and ADM are not implemented well. Between 2016 and 2019, the Robodebt scheme in Australia caused 470,000 incorrect debt notices to be sent by a computer system. This led to severe distress and even a few deaths. In 2012, Dutch Tax Authority used AI to detect fraud in childcare allowances. This disproportionately targeted low-income households and individuals with a foreign or dual nationality. Over 35,000 parents saw their subsidies abruptly and incorrectly cut, with more than a thousand children placed into state care and separated from their parents. Here in New Zealand, systems like Novopay demonstrate how our government's use of automation can go wrong.
In addition to these concerns about accuracy, public sector staff often have strong relationships with Māori, rural communities, and vulnerable populations. Automated decision-making cannot replace the expertise, nuance, and warmth that human relationships bring to our public service. In some cases, outsourcing the relationships to overseas machines may remove the very thing that makes those programmes effective.
The examples above show that AI and automated decision-making do not always make a process better. Sometimes, they introduce errors and harm to the people involved. It could bake in bias, compromise privacy, and erode the fundamental trust between the people and the state. Sometimes, they remove the most important element - the relationship, the institutional knowledge, the human compassion - and risk making the service worse, rather than better.
Currently, the public sector relies on voluntary frameworks like the Algorithm Charter for Aotearoa New Zealand. These are well-intentioned, but voluntary guidelines are no longer enough to manage the complex, high-stakes realities of modern machine learning and generative AI models. We need new laws with enforceable accountability to ensure that AI is working for us, not happening to us.