Two Australian stories landed on the same day this week, and together they describe the deal being offered to workers right now: regulate the technology at the frontier, outsource the work underneath it.
The regulator
Australia announced it will legislate mandatory AI safety rules, enforced by a new regulator with penalty powers, after an OpenAI agent was found to have accessed the Medicare statistics portal. The government's framing is obligations rather than aspirations: breach reporting, human oversight for high-risk uses, and penalties for companies that ship without them.
That is a real shift. Until now the Australian position was voluntary guidance plus a safety institute that advised. A regulator with the power to fine is a different kind of institution, and the trigger matters more than the announcement - a documented case of an autonomous agent reaching data it should not have been able to reach.
The awkward part is scope. A regulator aimed at frontier model developers does not touch the ordinary corporate decisions where AI shows up as a cost line. Which is where the second story starts.
The jobs
Qantas is cutting about 850 Australian corporate roles - roughly 3 per cent of its 28,000 local workforce - and moving the work to Accenture, with most of it delivered by teams in India. Finance, procurement, HR and marketing go first. The first round lands by mid-2027.
The statement says the services will be "further transformed through automation and AI over time". Read that twice. The roles are not being replaced by AI this year. They are being moved to a vendor whose contract promises automation later, and the headcount leaves either way.
The union called it a broken promise on offshoring. The company called it modernisation. Both are describing the same line item.
Why the two belong together
The functions being moved share a profile: process-heavy, documented, measurable. That is what makes them easy to hand over, and it is also the profile that current AI tooling handles best. So the outsourcing deal is not an AI story yet - it is a labour arbitrage story with an AI roadmap attached - but the roadmap is the part that removes the remaining roles in the second contract term.
Meanwhile the regulator being created covers model developers. It does not cover a company reorganising its back office around model output. The oversight and the disruption are pointed at different actors, which is the gap worth watching over the next two years.
If this reaches your company, three tells
The vendor announcement arrives before the automation. The saving is counted in year one, the automation lands in year three, the headcount never comes back.
The function list is the tell. Finance, procurement, HR and marketing are the areas where good documentation makes handover possible. If your team's work is fully written down, it is portable - to a vendor, or to a model.
The long notice period is a staffing requirement, not kindness. Eighteen months exists because someone has to run the process while the systems are replaced.
What to actually do with this
If you are on the receiving end, the useful question is not "will AI take my job" - it is "is my work documented well enough to be handed to someone else". If the answer is yes, start building the part that is not: judgement on edge cases, the relationships, the decisions nobody wrote a process for.
If you are building the automation, notice which side of the line you are on. The vendor gets the contract; the 850 people get the notice period.
The counter-argument, fairly stated
There is a defensible version of the Qantas decision. The airline is running legacy back-end systems, thin margins and a fuel bill that is going up; it also used the same day to flag fare rises and schedule cuts. Moving 850 roles to a vendor takes cost out in a way customers never see, which is exactly what a carrier in that position does.
The counter-argument to that is not that the airline should keep the roles forever. It is that the AI framing does the political work while the automation delivers the cost, and nobody has said how many people the vendor will need once the systems have been replaced. The honest version of the announcement would separate the two: "we are outsourcing this work" is a decision that can be argued with. "We are adapting to AI" ends the argument before it starts.
What the regulator will and will not catch
Three limits are worth knowing, because they will shape the next round of stories.
It covers developers, not deployers. The obligations sit with the companies building frontier models. A bank that automates loan decisions with a vendor's API is governed by financial regulation, not by AI safety law.
Breach reporting needs a breach people can see. The Medicare portal access became a case because it was documented and reported. Silent failures inside corporate workflows - the ones that move data, drop leads and misfile invoices - rarely produce a report.
Penalties are a deterrent, not a remedy. A fine does not restore the 850 jobs, and it does not tell the person whose role moved to a vendor what to do next.
The gap between those two is where most of the next two years of AI news will happen: companies making ordinary cost decisions under an AI banner, and a regulator built for a different problem.
The number that matters
850 roles is 3 per cent of 28,000. That is small enough to read as routine and large enough to matter to every person in it. Multiply it across the corporate sector and you get the shape of the next five years of white-collar work: not a wave of AI replacing people on a Tuesday, but a steady reclassification of documented work as vendor work, with the automation arriving on a schedule nobody published.
Full write-ups, with the numbers and the questions people are asking: Australia built an AI regulator after an agent broke into Medicare and Qantas cuts 850 jobs and hands the work to Accenture in India.
If you run automation that nobody is watching, the companion piece is how to catch an n8n workflow that stopped without failing.
Originally published at voxahq.in.
Top comments (0)