Let's start with the good news, because there is some, and we do not get it handed often enough.
New research surveyed workers in four countries about AI at work: Australia, Canada, New Zealand, and the United Kingdom. Of the four, New Zealanders are the least worried that AI is making them replaceable, and carry the least guilt about using it. Not marginally. Lowest on both counts.
If that produces a quiet moment of national satisfaction, we should take it. The rest of the world is busy having an argument about AI and fearing the end of work, while the Kiwi workforce has looked at the same technology and largely decided it will be fine. That is not naivety. It is the same steadiness that shows up whenever the sky is reportedly falling, and it is worth more than it sounds — a workforce that is not frightened of a tool is a workforce that will actually learn to use it.
So we have the temperament. What we do not yet have is the plumbing.
Because the same survey found something else. Somewhere in your business this morning, someone opened a chatbot in a browser tab they will close before anyone walks past. They pasted in a draft. They took what came back, tidied it, and sent it on under their own name. They are good at their job. They are not trying to deceive anyone. And they are fairly sure they should not mention it.
Roughly a third of the New Zealand workforce is doing some version of that.
The numbers, both halves of them
Employment Hero's AI Paradox at Work report surveyed 3,290 business leaders and 5,454 workers across Australia, Canada, New Zealand, and the United Kingdom, with a New Zealand sample of 256 business leaders and 820 workers, conducted with Focaldata between 23 April and 7 May 2026.
Two things are worth knowing before we get to the figures. The New Zealand sample covers businesses with between five and 1,500 staff, so it does not speak for sole traders or for the very largest employers. And at 820 workers, the margin of error runs to roughly three percentage points either way — enough to trust the shape of the findings, not enough to argue over a single point.
First, the flattering part, in full. Only 34 per cent of New Zealand workers feel AI is making them more replaceable — the lowest figure of any market surveyed. The guilt figure is likewise the lowest of the four. And 57 per cent say AI is helping them build more valuable skills, while 51 per cent have taught themselves AI through social media, on their own time, without being asked.
That last number deserves a second look. Half the workforce has gone and learned this by themselves, off YouTube and TikTok, because they wanted to. New Zealand workers are also, unlike every other market surveyed, more inclined to say their employer should be leading that development rather than leaving it to them.
Read those together and you have something most countries would like to have: a calm, curious, self-starting workforce, asking to be shown how to do this properly.
Now the other side of the ledger.
Thirty-two per cent of New Zealand workers admit to presenting AI-generated work as their own, without disclosing it. Twenty-eight per cent say they use AI tools without their employer knowing. Thirty-eight per cent say using AI to complete parts of their job feels like cheating, and a similar proportion report a sense of guilt when they use it to produce good work.
Put those percentages in a room. In a team of ten, three are handing over work they did not entirely write and have decided not to mention it. Nearly three more believe that what they are doing amounts to cheating.
And the detail that should stop you: the workers most competent with AI are the most conflicted about using it.
Hold those two halves together, because the combination is the whole story. This is not a frightened workforce hiding from a threat. It is the most relaxed, most self-motivated workforce in the study, concealing anyway.
Which tells you the concealment is not coming from fear. It is coming from not being told.
The obvious reading, and why it is not enough
The report's own conclusion is that New Zealand businesses should move faster. Adoption here is low by international standards: only 15 per cent of New Zealand SMEs describe AI as core to their operations, and 34 per cent of employers say it is driving innovation, against 45 per cent in Australia and 55 per cent in the UK.
That reading is not wrong. It is also, it should be said, the reading you would expect from a company that sells an AI-powered employment platform. Adopt faster is a fine message when adoption is what you are selling.
But set the adoption question aside for a moment and look at the two numbers side by side. Low organisational adoption. High individual concealment.
That combination is worse than either figure alone, and it is worse than high adoption would be. An employer with visible, governed AI use has something to point at: a policy, a tool list, a record of who used what and when. An employer whose staff are quietly using consumer chatbots has all of the same exposure and none of the record.
The risk is not that your people are using AI. It is that you cannot see any of it, and neither can anyone else who might later ask.
Where this lands in employment law
The Employment Relations Authority has already begun answering the question this data raises.
In Rennie v DJ Auto Vehicle Importer Ltd [2026] NZERA 27, an employee used ChatGPT to help draft a CV. The result contained inaccuracies — employment that had not happened, a dismissal left out. The Authority was untroubled by the involvement of the tool. Responsibility for the content sat with the person who submitted it, whatever had produced the words.
That principle is easy to accept when it runs against an employee. But it runs the other way too.
If a manager uses a chatbot to draft an investigation report, a warning letter, or a set of restructuring criteria, the employer owns what it says. Not the vendor. Not the model. The employer. And under section 103A of the Employment Relations Act, the question a member will ask is what a fair and reasonable employer could have done in the circumstances, which quickly becomes a question about what actually happened, who checked it, and whether anyone can show that now.
That is where undisclosed use becomes expensive. Twenty-eight per cent of your workforce using tools you do not know about means twenty-eight per cent of your work product has no verification trail behind it. If a disciplinary decision, a performance assessment, or a redundancy scoring exercise turns out to have been drafted by a system nobody documented, the employer is not defending the AI. It is trying to reconstruct its own process after the fact, without records.
There is a privacy dimension as well. Information pasted into a public AI tool has left the organisation. Where that information includes employee data — a complaint, a medical note, a performance file — the obligations under the Privacy Act 2020 do not pause because the disclosure was well intentioned or invisible.
