AINEWS
The hidden AI worker: why Aussie employees are using tools their bosses don't know about
A survey of 2,600 Australian workers finds three quarters say AI has improved their productivity — yet one in three keep it secret from their employer.
A survey of 2,600 Australian workers finds three quarters say AI has improved their productivity — yet one in three keep it secret from their employer.
Roughly one in three Australian workers use AI tools at work without their employer's knowledge. The figure comes from Employment Hero's new "AI Paradox at Work" survey of more than 2,600 Australians, and the same survey reports that about three-quarters of those workers say AI has improved their productivity.
Roughly two in five workers feel guilty about using AI at work, and many describe the experience as feeling like cheating. Employment Hero's chief executive, James Keene, framed it this way: "surreptitiously employees have moved ahead of their employers."
Workers are filling the vacuum themselves, and how they fill it is what makes the risk concrete. The survey found many are teaching themselves AI through TikTok, YouTube, and blog posts rather than employer-provided training, with no quality control on the material.
The National AI Centre's December 2025 to February 2026 adoption insights found roughly half of Australian organisations are now adopting AI, but 19% of small and medium businesses say they do not know how to use it, up two percentage points on the previous quarter. The Australian Bureau of Statistics' Business Characteristics Survey for 2024-25 found 46% of Australian businesses are already finding savings or efficiency gains through innovation that includes AI. The upside is real, but the smallest operators are the ones still fumbling for a starting point.
The cost of employer silence is not just secrecy. Clinical psychologist Dr Anna Kiaos, who works on workplace stress, has pointed out that unclear AI guidance is itself a source of anxiety. Workers worry the tool will replace them, or that work produced with AI will be judged as lesser. The guilt the survey picks up is the visible surface of a quieter fear.
Keene's constructive pivot is worth taking seriously. He argues employers can break the cycle by setting clear, visible AI guidance and modelling the behaviour themselves. That means writing an AI policy staff can read, showing their own AI use in meetings, and treating the tools as ordinary workplace software, rather than defaulting to silence or surveillance. Workers are not hiding AI because they want to break the rules. They are hiding it because the rules have not been written.
The catch is what the survey cannot measure. The productivity and quality gains are worker self-reports, not employer-validated output. Independent studies of generative AI in knowledge work have repeatedly found real productivity gains in narrow tasks, but the magnitudes vary and depend on what workers are being asked to do. Treating the 75% and 74% as anything more than a perception signal would over-read the poll. That workers believe AI is making them better is itself the story; whether they are right is a separate question the survey was not designed to settle.
The next data point to watch is whether the SME figure in the National AI Centre's adoption insights moves. If the 19% of small and medium businesses that say they do not know where to start keeps climbing, the hidden-use number will climb with it.