AI monitoring at work: what Australian staff should know
A friend who manages a call centre in outer Melbourne mentioned, almost in passing, that her team’s software now scores agents on “tone” during every call, using an AI model neither she nor her staff have ever seen the inside of. Nobody signed anything new. It was just switched on one Tuesday, folded into an existing software update. That’s the shape AI workplace monitoring usually takes in Australia right now: quiet, incremental, and bundled into tools staff already use.
We’ve been getting more reader emails about this than almost any other AI topic this year, which tells us something: the anxiety isn’t about chatbots writing emails, it’s about being watched by one while you work. Fair enough. The gap between what employers are legally allowed to do and what workers assume they’re allowed to do is wide, and it’s mostly unexamined.
What’s actually being monitored
Strip away the marketing language and workplace AI monitoring tends to fall into a few buckets. There’s activity tracking: keystrokes, mouse movement, app usage, idle time, often dressed up as a “productivity” or “focus” score. There’s communications analysis: AI scanning emails, Slack or Teams messages for sentiment, policy breaches or “flight risk” language. There’s call and meeting analysis, increasingly common in retail and finance call centres, where AI transcribes and scores tone, compliance phrasing and even talk-to-listen ratios. And there’s location and movement tracking, common in logistics and delivery work, sometimes extending to wearables that track physical exertion or fatigue.
None of this is new in principle. Call centres have recorded calls “for quality and training purposes” since landlines were the only option. What’s changed is scale and interpretation: a human supervisor used to spot-check a handful of calls a week. An AI model now reads every single interaction, scores it, and feeds that score into performance reviews, sometimes without a human ever listening in. That’s the actual shift worth paying attention to, not the existence of monitoring itself.
Our take: the technology isn’t the problem so much as the quiet removal of human judgement from decisions that affect someone’s pay packet or job security.

What Australian privacy law actually covers
This is where a lot of the anxiety and a lot of the misinformation collide. The Privacy Act, overseen by the Office of the Australian Information Commissioner, applies to most organisations, but there’s a long-standing exemption for employee records held by private sector employers, when that information is used for a directly employment-related purpose. In plain English: your employer collecting and using data about your work performance, through monitoring software, generally sits outside the Privacy Act’s usual protections, provided it genuinely relates to the employment relationship.
That exemption surprises people every time we mention it. It means the usual avenues, like lodging a complaint with the OAIC about how your employer handles your data, often won’t apply to garden-variety workplace monitoring. Surveillance at work is instead governed by a patchwork of state and territory surveillance laws, which vary depending on whether you’re in New South Wales, Victoria or elsewhere, plus the Fair Work Act’s general protections, which can apply if monitoring is used to punish someone for a protected reason such as raising a workplace complaint. The Fair Work Ombudsman is the right first stop for that category of dispute, not a privacy regulator.
Notification still matters
Most state surveillance laws require an employer to notify staff, in writing, before introducing monitoring of computer use, and in many cases before using cameras or tracking devices. Covert monitoring without notice is generally unlawful outside narrow, specific circumstances, usually tied to suspected unlawful activity and often requiring a magistrate’s authorisation. If your employer rolled out AI scoring without telling anyone, that’s worth raising, even if the monitoring itself turns out to be permitted once disclosed properly.
The consent question nobody quite answers cleanly
Here’s the mildly unpopular view we’ll actually defend: “consent” in an employment relationship is mostly theatre, and pretending otherwise doesn’t help workers. When a company says staff have “consented” to AI monitoring because they signed an updated policy or clicked through a notification, that’s consent in the thinnest possible sense. Nobody realistically declines and keeps their job. We’d rather employers be upfront that this is a condition of employment, not a choice, because at least that’s honest about the power imbalance. Dressing it up as a free choice, which a lot of corporate communications teams do, is the bit that actually erodes trust.
That doesn’t mean monitoring is automatically wrong. It means the “you agreed to this” framing deserves more scepticism from workers than it usually gets, and a bit more honesty from employers than it usually receives.
Where this echoes an older argument
Workplace surveillance panic isn’t new. Factory time-and-motion studies a century ago provoked almost identical objections: that measuring every movement strips workers of autonomy and treats people as inputs to be optimised. AI monitoring is the same argument with better data and worse transparency, because at least a stopwatch-wielding supervisor was a visible, named person you could push back against. An AI model scoring your tone of voice is harder to argue with, because there’s no one in the room to argue with. The accountability gap, not the measurement itself, is the real continuity from that era to this one.
Retail, logistics and call centres are the front line
If you want to see where this is most advanced in Australia, look at large retail and logistics operations and outsourced call centres, where margins are thin and headcount costs are closely watched. Delivery driver apps tracking routes and idle time, warehouse scanners logging pick rates to the second, and AI call scoring in telco and insurance call centres are all established practice rather than emerging trends. Office workers using Microsoft 365 or Google Workspace are catching up fast too: both platforms now include analytics features that can surface individual activity patterns to managers, even if neither company markets them primarily as surveillance tools.
We’d also flag the home-working angle, since plenty of Australians working from a spare room in Parramatta or a shared desk in a Geelong co-working space assume they’re less visible than someone in an open-plan office. Often the opposite is true: remote workers tend to be monitored more closely precisely because a manager can’t glance across the floor to see who’s at their desk. If you went remote partly to escape oversight, the irony is that the software stepped in where the manager’s eyeline used to be.
What you can actually do about it
A few practical, non-dramatic steps. First, read the policy. Most organisations that introduce monitoring are required to document it somewhere, often in an IT acceptable use policy or an enterprise agreement; knowing exactly what’s measured is more useful than assuming the worst. Second, separate personal and work activity properly: personal email, personal banking and personal messaging on a work device or work network is visible more often than people expect, AI monitoring or not. Third, if you believe monitoring has been introduced without proper notice, or used to target you after a workplace complaint, the Fair Work Ombudsman and your union, if you have one, are the right first call, not a privacy complaint that likely won’t apply.
If you’re setting up any of this on the employer side, or just trying to understand what a work laptop can see versus what your personal phone can see, it’s worth separating the two environments properly; our piece on desktop vs laptop for working from home covers some of the practical boundary-setting involved. And if your employer is rolling out AI tools more broadly, not just for monitoring, our rundown comparing ChatGPT vs Claude vs Gemini for Australians is a reasonable primer on what those systems can and can’t see of your data once you’re using them for work tasks.
The bit employers tend to get wrong
We’d argue the biggest mistake isn’t the monitoring itself, it’s treating an AI score as if it were a fact rather than an estimate with an error rate. A sentiment model misreading a stressed customer service call as “hostile”, or a keystroke tracker flagging someone reading a document carefully as “idle”, isn’t a hypothetical. These systems make mistakes constantly, and the mistake becomes a permanent data point in someone’s performance file unless a human actively overrides it. Any workplace using AI scoring without a clear, genuinely used process for staff to query and correct a score is building something closer to an unaccountable black box than a management tool. That’s a governance failure, not a technology one, and it’s entirely fixable if an organisation actually wants to fix it.
On the broader question of how much personal data these systems quietly accumulate even outside work, our explainer on the privacy cost of on-device AI is a useful companion read, since a lot of the same “it’s just background processing” logic gets applied in both settings.
What this means for your next job interview, too
Worth remembering this cuts both ways before you’ve even started a job. AI tools are now commonly used to screen résumés, analyse video interviews for “confidence” or “engagement” signals, and rank candidates before a human ever looks at the shortlist. The same scepticism about black-box scoring applies there. If you’re asked to do a one-way video interview scored by software you can’t see or question, it’s entirely reasonable to ask the recruiter, plainly, whether a human reviews every result or only the ones the algorithm flags as promising. Most won’t have a confident answer, and that tells you something too.
None of this means refusing to engage with AI-driven workplaces is realistic; it’s spreading too fast for that. But understanding the actual legal gaps, rather than the ones people assume exist, is the difference between a reasonable question to HR and a complaint that goes nowhere because it’s aimed at the wrong regulator.
— *Dale Whitfield, Founding Editor, Tech Geek*




