Australians are continuing to experiment with AI in the workplace, but a knowledge gap between leaders and frontline workers reveals governance risks.
Sarah Morton
Editorial Lead, Asia Pacific & Japan
Workday
In June 2026, rogue AI agents breached two government websites in Australia, pushing the conversation on AI governance to the top of the national agenda.
While no personal data was compromised, the incidents sparked a rapid review of AI systems in government and served as a warning about the risks of embracing AI adoption without sufficient guardrails in place.
Our latest AI@Work survey offers some insight into the factors that might be contributing to this governance risk.
Namely, that while Australians are embracing AI in their daily work, unclear rules and a lack of AI knowledge are creating a gap between ambition and readiness.
The AI@Work survey polls approximately 1,000 Australian employees each month, as part of a global study of 6,000 workers across Australia, Canada, France, Germany, the UK, and the US.
Report
According to the September AI@Work survey, 80.1% of Australian respondents report using AI for at least some of their work, outpacing the global average of 77.5%.
More significantly, 65.4% of Australian AI users say they have already allowed an AI tool to complete a task fully autonomously at least once in the past month.
However, despite this widespread experimentation, a knowledge gap exists between those setting the strategy and those executing it.
The data shows that 65.2% of Australian senior leaders say they understand AI agents well or could confidently explain the term. In contrast, only 24.5% of Australian individual contributors say the same.
This 2.7x familiarity gap means that while executives are planning for a future driven by autonomous agents, the employees expected to use these tools daily lack the basic vocabulary and understanding to keep up.
If the people closest to day-to-day operations do not have the same clarity and support as senior leaders, AI will fail to become a true workforce advantage.
While recent government breaches underscore the need for robust governance of AI agents, the risk begins with unclear rules around usage.
Only 35.2% of Australian employees feel their employer's expectations for disclosing AI use are 'very or extremely clear'.
Globally, nearly one in five employees (19.2%) say their employer has no AI disclosure expectations at all.
When employees do not understand the rules, they either make poor delegation choices or they hide their usage.
In Australia, 38.9% of AI users admit to deliberately concealing their AI use on a work task.
While this is slightly lower than the global average of 41.3%—and significantly better than France (46.2%) and the US (45.1%)—it still represents a substantial portion of the workforce operating in the shadows.
This shadow agent risk means organisations lack visibility into how AI is being deployed, making it difficult to manage risk and maintain regulatory compliance at scale.
Only 40.7% of Australian workers agree that they can rely on AI outputs without close checking.
Despite these governance challenges, sentiment among Australian workers continues to be remarkably positive.
Australian employees report a net work-experience sentiment score of +51.8, leading the pack globally and sitting far above peers like Germany (+34.1) or France (+27.8).
But this optimism comes with a cost.
While AI is saving the average Australian user roughly three hours a week, a persistent 'rework tax' is eating into those gains.Almost a third (32.8%) of Australian AI users report losing at least half of the time AI saves them to fixing, reworking, or correcting low-quality outputs.
Furthermore, only 40.7% of Australian workers agree that they can rely on AI outputs without close checking.
While Australian rework is slightly lower than the high-friction US market (where a massive 41.6% report heavy rework), it remains a significant drag on genuine productivity.
It also illustrates the danger of operating without clear rules and enterprise-level guardrails.
As AI agents take on more autonomous work in Australian organisations, this rework tax has the potential to transform from a simple time-waster into a serious governance risk.
As organisations adapt to the new rules of the governance game, recent headlines are a reminder that organisational risk starts and ends with people.
AI@Work results show that only 56% of Australian workers have had a manager discuss how AI will impact their role in the past month.
Without active guidance from leaders, employees are left to navigate tool selection and task delegation on their own.
This disconnect was also visible in our recent Global Workforce Report, The Adaptability Advantage.
It found that while 79% of workers said they know what skills they need to succeed, only 66% felt their employer actually supports them in developing those skills.
Building a workforce that understands and is familiar with AI agents, discloses their AI use, and has access to tools that limit AI rework is critical to ensure that organisations can manage risk as adoption continues to scale.
Organisations must establish clear expectations for human oversight.
Here’s what else we’re tracking month over month. Come back in November for new insights and trends.
| KPI | What It Measures | August Results | September Results | MoM Change |
|---|---|---|---|---|
| AI Adoption Rate | % of employees using AI for any part of their work, excluding “none of it” and “I don’t use AI at work” | 75.7% | 80.1% | +4.3 pts |
| Power-User / Heavy-Use Rate | % using AI for “most” or “almost all” of their work | 17.7% | 19.3% | +1.7 pts |
| Agentic Task Delegation Rate | % of AI users who have let an AI tool complete a task fully autonomously with little or no input at least once | 63.0% | 65.4% | +2.4 pts |
| Average Weekly Time Savings | Mean hours per week employees say AI saves them, using bucket midpoints | ~2.8 hours/week | ~2.9 hours/week | +0.1 hrs |
| AI “Rework Tax” | % of AI users who say 50%+ of their time savings is eaten up fixing or reworking low-quality AI output | 26.8% | 32.8% | +6.0 pts |
| Trust in AI Output Accuracy | % who agree or strongly agree they can rely on AI outputs without close checking | 40.0% | 40.7% | +0.7 pts |
| Shadow AI / Concealment Rate | % of AI users who deliberately hid their AI use on a work task | 37.1% | 38.9% | +1.8 pts |
| AI + Work Experience Sentiment | Net score: % saying AI made day-to-day work better minus % saying it made it worse | +51.3 pts | +51.8 pts | +0.6 pts |
| Employer AI-Trust Score | % who somewhat or strongly trust their employer to use AI in a way that benefits employees, not just the organisation | 46.4% | 46.9% | +0.4 pts |
| Career Confidence Gap | Net career optimism: % agreeing AI will create new opportunities minus % agreeing AI will make current skills less valuable | +3.8 pts | +6.0 pts | +2.2 pts |
| Training | % of employees who have received AI training from their employer | 76.1% | 79.4% | +3.2 pts |
| Manager AI Conversation Rate | Share with a manager who discussed AI role impact in past month | n/a | 56.0% | n/a |
Source: Workday August and September 2026 AI@Work surveys. Results shown are for Australia.
Report
Why task-oriented AI is failing the enterprise in Australia and New Zealand.