AI Jargon, Explained

The terms behind the headlines.

AI reporting is full of words that get thrown around without ever being explained. This page breaks them down in plain English, with the New Zealand context that usually gets left out. New terms are added regularly. Follow @aisafetyaotearoa on Instagram for the latest.

Shadow AI Black Box Hallucination
01

Shadow AI

/ˈʃædoʊ eɪ aɪ/

What it means

When people secretly use unapproved AI tools without anyone else's knowledge or oversight. From corporate offices to classrooms, New Zealanders are quietly turning to unvetted chatbots to clear work backlogs and get through assignments under pressure.

Why it matters

Most people forget that AI chatbots do not just read the information you paste into them. They absorb it. Paste a client's financial details into a public AI tool to save ten minutes, and that information may be stored, used to train the model, or surface somewhere you never intended. The convenience is real. So is the exposure.

In New Zealand

A 2023 ISACA survey found 63% of employees in Australia and New Zealand were already using AI at work, while only 11% of their organisations had a formal policy governing how. That gap has not closed. Research from Kordia in early 2026 found 24% of New Zealand businesses now rank staff misuse of AI among their top three cybersecurity challenges, up from 16% the year before. The pattern is consistent. New Zealanders are adopting AI faster than their workplaces and schools are building the guardrails to govern it.

02

Black Box

/ˌblæk ˈbɒks/

What it means

A system where you can see the input and the output, but not what happens in between. Some AI systems are built from networks so large and complex that no single person fully understands how they reach a given answer, not even the people who built them.

Why it matters

In New Zealand, AI is already being used to help assess hiring decisions, insurance applications, and loan approvals. When a black box system is involved, neither the applicant nor, in some cases, the organisation using the tool, can fully explain why it reached the conclusion it did. Knowing how those decisions are made matters when the outcome affects your job, your cover, or your finances.

In New Zealand

The real danger in New Zealand is a legal blind spot. Because we lack a specific "right to an explanation" law for AI, organisations can easily drift into a culture of "computer says no". If a black box algorithm incorrectly flags your credit history, replicates systemic racial or gender bias in a hiring pool, or unfairly spikes your insurance premiums, the inner workings remain hidden. This creates a massive accountability gap: every day Kiwis risk being locked out of vital jobs, housing, and financial services by an invisible, uncheckable system that even the companies themselves cannot fully explain.

03

Hallucination

/həˌluːsɪˈneɪʃən/

What it means

When an AI confidently states something that is completely made up. Not a glitch, not an admission of uncertainty. The system generates plausible sounding information that does not exist and presents it with exactly the same confidence as something true.

Why it matters

New Zealanders are increasingly turning to AI tools for legal advice, medical questions, and financial decisions. When the AI gets it wrong, it will not flag the error. The false information reads exactly like the true information.

In New Zealand

This is not theoretical. In 2026, the Supreme Court dismissed a case brought by a self-represented litigant whose filings cited legal authorities that did not exist, including one called "Peterson v Forbes,". The Justices wrote that many of the litigant's authorities were either irrelevant or did not exist, appearing to be hallucinations of a generative AI application. It was not an isolated incident. Earlier the same year, in a separate Supreme Court decision, the Court warned that unsupervised use of AI in court submissions can amount to contempt of court. New Zealand courts have since flagged the same problem in the family court, the employment court, and the Employment Relations Authority.