AI risks unite parties – but not on solutions

A new AI Election Scorecard reveals political parties agree advanced AI carries serious potential risk but remain poles apart in terms of how to address these risks.

Good Ancestors New Zealand released its AI Election Scorecard today, which shows AI has become a real issue in this election, with parties setting out how they would manage the risks advanced AI poses to New Zealanders and national security.

The not-for-profit organisation rated seven parties on six policy areas, analysing published policy, public statements and responses to survey questions. Only Labour, the Green Party and Opportunity have published substantive AI policy ahead of the election.

The Green Party, National, Opportunity and Te Pāti Māori answered the survey. Good Ancestors New Zealand has invited Labour, ACT and New Zealand First to respond before election day.

Good Ancestors Executive Director Dylan Chambers says it is pleasing the risks are being recognised, but the solutions are not going far enough to address them.

“We need to be taking the next step of making institutional and policy change to manage those risks. Several parties have taken positive steps, but most have not gone as far as we’d like to see.

“AI is becoming more capable faster than governments are building the ability to understand or control it. Top of the list should be establishing an AI Safety Institute, which would give New Zealand dedicated capability to evaluate these systems, connect with our global partners and advise Ministers on the risks that are coming.”

Chambers says New Zealand also has no requirement for AI companies to report serious incidents, and it is disappointing that only one party has a formal commitment to fix that.

“This is a gap the next government should fix. We saw the impact of no formal reporting recently in Australia where an OpenAI model hacked a government health website in June and the company did not notify the government of the breach for some months after. Reporting must be law.”

Chambers says access to leading models also needs to be treated as a strategic issue, and it’s not enough to assume that New Zealand will have access.

“Leading AI models are becoming core economic infrastructure and a core part of our cyber defences, and New Zealand builds none of its own. In June we were cut off from one of the most capable models with no notice for almost three weeks. The next government needs to plan for retaining access as it already does for other key economic inputs like fuel.”

The scorecard rates each party on six questions: whether it recognises advanced AI as a serious risk, and whether it will lift government capability, strengthen AI leadership and strategy, update laws, keep New Zealand’s access to leading AI models and require reporting of serious AI incidents.

A formal commitment is a clear public commitment made in published party policy or in a public statement on the party’s behalf. A qualified or informal commitment is one made only in a party’s survey answers, or a public position that comes close to the commitment but falls short. Parties are listed in order of their overall score.

The key findings for each party from the AI Election Scorecard include:

Opportunity: Has formally committed to four of the six policy areas, and is the only party formally committed to mandatory reporting of serious AI incidents or to maintaining New Zealand’s access to leading AI models. It recognises advanced AI risks and commits to updating laws. It has commitments on government capability and AI leadership, but not to an AI Office, an AI Safety Institute or a Minister for AI.

Labour: Has formally committed to four of the policy areas and is the only party formally committed to a Minister for AI. It also backs establishing government capability through an Office of AI within DPMC and publishing a cross-government programme of work, recognising advanced AI risks and updating laws. Labour has proposed an online safety regulator for AI-enabled harms, but has not committed to requiring reporting of serious AI incidents, and it has no public position on access to advanced AI models.

Green Party: Formally backs lifting government capability with an AI Office within DPMC and law changes. Its published AI policy treats unregulated AI as a significant risk to society and the economy, but does not directly address the most significant risks to public safety and national security. The party answered yes in the survey to appointing a minister with cross-government responsibility for AI and publishing a whole-of-government plan, but these positions do not feature in its policy announcements.

Te Pāti Māori: Answered yes to every survey question, acknowledging AI risks and supporting an AI Safety Institute, a Minister for AI to lead a comprehensive plan, legal review and serious-incident reporting. However, its answer on retaining access to leading AI models described allowing continued access to leading AI models under regulation and standards, not acting to secure it. None of these positions is yet backed by a formal public commitment or policy announcement.

National: Answered the survey and clearly recognises advanced AI risks, including to national security. In government, it agreed a Five Country statement committing to enabling timely access to frontier AI models. But on government capability, AI leadership and strategy, law and incident reporting, National would rely on existing agencies and change settings only once gaps emerge. That is a reactive approach that waits for harms to appear, and none of those agencies has the job of assessing advanced AI risks.

ACT: No formal commitments. Informally recognises advanced AI risks but offers only partial support for dedicated government capability and law changes. It has no public position on a Minister for AI, strategy or access to advanced models, and on incident reporting it says existing guidance and regulators are enough. It has not yet answered the survey.

New Zealand First: Has called for caution on AI and stronger safeguards against digital exploitation, but has not clearly acknowledged risks to public safety or national security. It supports updating some laws to deal with deepfakes, but has not committed to a wider review of the law. No public position was found on the other four areas, and it has not yet answered the survey.

See the full scorecard at goodancestors.org.nz/scorecard.