The Government has started a four month trial of a generative artificial intelligence tool to help write import health standards, the legal documents that decide which plants, animals and biological products are allowed across the New Zealand border and on what terms.
Biosecurity Minister Andrew Hoggard announced the pilot on 25 July, presenting it as a fix for a standards process he says has fallen behind the pace of modern trade. “Import health standards set the biosecurity rules goods must meet before entering New Zealand. They manage risks from plant, animal, and biological products,” he said. “The way we develop these standards has not kept pace with modern trade and innovation. We are improving the system so it can move safely and faster, without compromising biosecurity.”
The Minister was careful to draw a line between drafting and deciding. “This is about using smart tools to take on document-heavy work, while key decisions remain with Biosecurity New Zealand experts,” he said. “The tool could free up staff time without cutting corners, with a four-month pilot underway to develop and trial the tool.” The trial, he said, “will test whether the tool can help deliver high-quality, evidence-based standards more safely and consistently.”
Import health standards are not guidance notes that officials can revise on a whim. They are legal documents issued under section 24A of the Biosecurity Act 1993, and they cover a long list of goods, from nursery stock and animal products through to shipping containers, used vehicles and tyres. The Ministry for Primary Industries builds each one in stages, starting with a risk analysis, then a draft that goes out for consultation, then a provisional version, then the final standard. Section 23 of the same Act requires the ministry to consult interested parties, including any affected government departments, before a standard is issued or amended, and those who disagree can ask for an independent review and even suggest who should sit on the review panel.
All of that generates paperwork, and the paperwork is what the Government wants a language model to help carry. Each standard rests on a pest risk analysis, a written justification for the requirements it imposes, and a consultation record that has to stand up if an importer or an exporting country challenges it.
The idea has been in the pipeline for a year. In July 2025 Hoggard launched a regulatory efficiency programme aimed at shortening the time it takes to develop a standard, saying the framework was more than 30 years old and struggling to adapt to fast-moving global trade. He said then that early testing had already been done. “We’ve already seen successful proof-of-concept trials that show AI can significantly reduce the time it takes to complete key parts of the IHS process,” he said in that announcement. “This is about using smart tools to do the heavy lifting, so our experts can focus on the decisions that matter most.” The announcement last weekend is the point where that concept becomes a working pilot inside Biosecurity New Zealand.
Hoggard tied the reform to farm incomes, noting that export revenue from the food and fibre sector is forecast to reach $64.3 billion in the year to 30 June 2026. Faster standards mean growers and nurseries get access to new plant material sooner, which matters for productivity and for adapting orchards and crops to a changing climate. A parallel piece of work is under way on the live plant import system, with the plant nursery stock pathway as the test case.
The cost of getting one of these documents wrong is what sits on the other side of the ledger. When Mycoplasma bovis was found in New Zealand cattle, the eradication programme agreed with DairyNZ and Beef + Lamb New Zealand was budgeted at about $886 million over ten years, with the Crown carrying 68 percent of it. No import standard is directly blamed for that arrival, but it is a fair measure of what a single organism can cost once it is established here.
Which raises the obvious question about what happens when a language model gets something wrong in a document that carries legal force. The Government’s own Responsible AI Guidance for the Public Service, published in February 2025, is blunt on the point. It describes hallucinations as cases where generative AI systems create incorrect yet convincing outputs, tells agencies to verify outputs against trusted sources before using them, and says generated content should never be treated as authoritative. Applied to an import health standard, that means a specialist still has to check every claim the tool produces, which puts a natural ceiling on how much time the pilot can actually save.
Not everyone is convinced the guidance is strong enough. In May, public management researchers Deborah Te Kawa and Barbara Allen, writing for The Conversation in a piece republished by RNZ, called the public service AI framework a “Pollyanna policy” that relies on good intentions without enforcement. “Principles without a legislative mandate become aspirational without accountability,” they wrote, arguing the non-binding approach pushes responsibility onto individual agencies whose capability to handle it varies widely.
The announcement leaves a fair bit unsaid. It does not name the AI system being used, say where it runs or who supplies it, or state whether draft standards produced with the tool’s help will be identified as such when they go out for public consultation. It also does not say what test the pilot has to pass at the end of the four months before the tool is used on standards that importers and trading partners will have to live by.
Both threads here, how quickly the border should clear new goods and how far a government should let AI into its own decision paperwork, are heading into the campaign. The Greens spent last week calling for a one-year pause on new AI data centre consents, an argument the Prime Minister dismissed as alarmist, and the parties are now setting out positions on where machines belong in public administration. The general election is on 7 November. If you want to compare where the parties stand on questions like these, our voting tool lets you work through the policies side by side.
Would you trust a drafting tool to help write the rules that keep pests out of the country, so long as a human signs them off? Tell us what you think in the comments below.
This article was written by AI, briefed to report the facts, hopefully without some of the bias people bring to the job 🙂
Leave a Reply