Canberra has become one of the busiest testing grounds in the country for artificial intelligence that watches, scores and flags human behaviour, and the question now being asked is a deceptively simple one: before any algorithm is allowed to punish a person, should a human being have to sign off first? Writing in the Canberra Times, the argument runs that the capital’s emerging rules for AI monitoring are missing the single safeguard that matters most, a guaranteed right of appeal to a person before a machine’s judgment becomes a penalty.
The context is familiar to anyone who has followed the spread of automated systems into Australian public life. Monitoring tools are no longer confined to counting keystrokes or logging when a pass is swiped at a door. They now estimate productivity, infer attention, detect anomalies in spending or attendance, and in some cases recommend consequences, from a flagged timesheet to a formal warning. Canberra, with its dense concentration of public servants, universities and research institutions, sees more of this technology in daily use than almost anywhere else in the country.
The news
The core of the argument is that monitoring rules being drafted and applied around the capital tend to focus on what the technology is allowed to collect, rather than on what it is allowed to do with the result. A system can be perfectly compliant with privacy obligations while still producing an outcome that lands on a worker or a student with real force, a docked score, a disciplinary trigger, a benefit withheld, without a person ever having looked at the case. The remedy proposed is blunt and old-fashioned: no algorithmic finding should convert into a punishment until a human has reviewed it and the affected person has had a genuine chance to contest it.
That principle sounds obvious, yet it is precisely the step that automated systems are built to skip. The efficiency case for AI monitoring rests on removing the human from the loop. A rule that puts the human back in, at the moment of consequence, cuts against the very reason many of these tools are purchased. The full argument is set out in the Canberra Times, which frames the appeal right as the line between oversight and automated punishment.
Two ways of seeing it
Supporters of a mandatory appeal see it as a basic due-process protection, the digital equivalent of the right to be heard before a decision goes against you. On this view, an algorithm is an input, not a verdict. It can surface a pattern, raise a flag or rank a risk, but the moment its output changes someone’s pay, record or standing, a person with authority and context should be the one making the call, and the subject should be told what triggered it and be able to put their side. Without that, they argue, the system quietly inverts a long-standing principle, forcing people to prove the machine wrong rather than requiring the institution to prove them at fault.
Those more sympathetic to automation push back on practical grounds. Mandating human review at every point of consequence, they say, reintroduces the cost, delay and inconsistency that the technology was meant to remove, and it can create a false comfort if the human reviewer simply rubber-stamps whatever the system recommends. The more useful version of the safeguard, in this reading, is not a human at every gate but transparency, auditability and a well-resourced appeal path, so that review is meaningful rather than ceremonial. The danger both camps acknowledge is automation bias, the well-documented tendency for people to defer to a machine’s output precisely because it looks objective.
What it means for Australia
Canberra matters here beyond its own borders, because what the public sector in the capital tolerates tends to set the template the rest of the country inherits. Commonwealth and territory agencies are among the largest employers in the nation, and the governance habits they adopt for workplace monitoring, student oversight and automated service decisions ripple outward into the practices that private employers and state governments later copy. The hard lessons of Robodebt still sit over this debate like a warning sign, a reminder of what happens when an automated process issues adverse outcomes at scale without adequate human checking or a workable way to contest them.
The appeal-first principle also speaks directly to a gap in Australia’s broader AI settings. The country has adopted voluntary AI safety guidance and is weighing mandatory guardrails for high-risk uses, and automated decision-making that affects a person’s livelihood or rights is squarely in that high-risk zone. A clear, enforceable rule that a human must review and an individual must be able to appeal before any algorithmic penalty takes effect would give that abstract policy ambition something concrete to stand on. It would also align Canberra with the direction of travel internationally, where a right to human review of consequential automated decisions is increasingly treated as a baseline rather than a luxury.
There is an equity dimension as well. The people most exposed to monitoring systems, casual and lower-paid workers, students under exam surveillance, recipients of government services, are often the least equipped to challenge an opaque decision on their own. An appeal right only protects them if it is easy to use, quick to resolve and not buried in process. A safeguard that exists on paper but is impractical in practice offers little to the person staring down a penalty they did not understand and were never asked about.
What’s next
The practical test is whether the principle is written into rules with teeth or left as an aspiration. That means spelling out which decisions count as consequential enough to require human sign-off, setting time limits so appeals are resolved before damage is done, and demanding that institutions disclose when an automated system has been involved in a decision at all. It also means measuring whether human reviewers are genuinely overturning flawed outputs or merely waving them through, because an appeal that never changes an outcome is not an appeal.
For now, the capital’s monitoring tools keep expanding faster than the governance around them, and the argument from Canberra is that the sequence should be reversed. Decide the safeguard first, the right to a human and a hearing before an algorithm can punish, and let the technology operate within it. Get that order wrong, and the efficiency gains arrive long before the protections do.
Sources: The Canberra Times.
